Tag Archives: machine learning

Soft Actor-Critic: Deep Reinforcement Learning for Robotics



Deep reinforcement learning (RL) provides the promise of fully automated learning of robotic behaviors directly from experience and interaction in the real world, due to its ability to process complex sensory input using general-purpose neural network representations. However, many existing RL algorithms require days or weeks (or more) worth of real-world data in order to converge to the desired behavior. Furthermore, such systems can be tough to deploy on complex robotic systems (such as legged robots) which can easily get damaged during the exploration phase, hyperparameter settings can be challenging to tune, and various safety considerations can introduce further limitations.

In collaboration with UC Berkeley, we recently released Soft Actor-Critic (SAC), a stable and efficient deep RL algorithm suitable for real-world robotic skill learning that is well-aligned with the requirements of robotic experimentation. Importantly, SAC is efficient enough to solve real-world robot tasks in only a handful of hours, and works on a variety of environments with a single set of hyperparameters. Below, we discuss some of the research behind SAC, and also describe some of our recent experiments.

Requirements for Real-World Robotic Learning
Real-world robotic experimentation brings significant challenges, such as constant interruptions in the data stream due to hardware failures and manual resets, and smooth exploration to avoid mechanical wear and tear on the robot, which set additional restrictions to both the algorithm and its implementation, including (but not limited to):
  • Good sample efficiency to lower the learning time
  • Minimal number of hyperparameters that require tuning
  • Reusing already collected data on different scenarios (known as off-policy learning)
  • Ensuring that learning and exploration does not damage the hardware
Soft Actor-Critic
Soft actor-critic is based on maximum entropy reinforcement learning, a framework that aims to both maximize the expected reward (which is the standard RL objective) and to maximize the policy's entropy. Policies with higher entropy are more random, which intuitively means that maximum entropy reinforcement learning prefers the most random policy that still achieves a high reward.

Why might this be desirable for robotic learning? The most obvious reason is that policies optimized for maximum entropy will be more robust: if the policy can tolerate highly random behavior during training, it is more likely to respond successfully to unexpected perturbations at test time. However, a more subtle reason is that training for maximum entropy can improve both the algorithm's robustness to hyperparameters and its sample efficiency (to learn more, see this BAIR blog post, and this tutorial).

Soft actor-critic maximizes the entropy augmented reward by learning a stochastic policy that maps states to actions and a Q-function that estimates the objective value of the current policy, optimizing them using approximate dynamic programming. In doing so, SAC views the objective as a grounded way to derive better reinforcement learning algorithms that perform consistently and are sample efficient enough to be applicable to real-world robotic applications. For technical details please see our technical report.

Performance of SAC
We evaluated SAC with two tasks: 1) quadrupedal walking with the Minitaur robot from Ghost Robotics, and 2) rotating a valve with a three finger Dynamixel Claw. Learning to walk presents a substantial challenge, as the robot is underactuated, and must therefore delicately balance contact forces on the legs to make forward progress. An untrained policy can lose balance and fall, and too many falls will eventually damage the robot, making sample-efficient learning essential.

Although we trained our policy only on flat terrain, we subsequently tested it on varied terrains and obstacles. In principle, policies learned with soft actor-critic should be robust to test-time perturbations, because they are trained to maximize entropy (i.e., inject maximal noise) at training-time. Indeed, we observe that the policies learned with our method are robust to these perturbations without any additional learning.
Illustration of learned walking, using SAC implemented on the Minitaur robot. A full video of the learning process can be found at our project website.
The manipulation task requires the hand to rotate a valve-like object so that the colored peg faces to the right, as shown below. This task is exceptionally challenging due to both the perception challenges and the need to control a hand with 9 degrees of freedom. In order to perceive the valve, the robot must use raw RGB images shown in the inset at the bottom right. The initial position of the valve is reset uniformly at random for each episode, forcing the policy to learn to use the raw RGB images to perceive the current valve orientation.
Soft actor-critic solves both of these tasks quickly: the Minitaur locomotion takes 2 hours, and the valve-turning task from image observations takes 20 hours. We also learned a policy for the valve-turning task without images by providing the actual valve position as an observation to the policy. Soft actor-critic can learn this easier version of the valve task in 3 hours. For comparison, prior work has used natural policy gradients to learn the same task without images in 7.4 hours.

Conclusion
Our work demonstrates that deep reinforcement learning based on maximum entropy framework can be applied to learn robot skills in challenging real-world settings. Since the policies are learned directly in the real world, they exhibit robustness to variations in the environment, which can be difficult to obtain otherwise. We also showed that we can learn directly from high-dimensional image observations, which represents a significant challenge in classical robotics. We hope that the release of SAC helps other research teams in their effort to adopt deep RL for more complex real-world tasks in the future.

For more technical details, please visit the BAIR blog post, or read an early preprint of the locomotion experiment and a more complete description of the algorithm. You can find the implementation on GitHub.

Acknowledgements
This research was done in collaboration between Google and UC Berkeley. We would like to thank all the people who were involved, including Sehoon Ha, Kristian Hartikainen, Jie Tan, George Tucker, Vincent Vanhoucke and Aurick Zhou.

Source: Google AI Blog


Exploring Quantum Neural Networks



Since its inception, the Google AI Quantum team has pushed to understand the role of quantum computing in machine learning. The existence of algorithms with provable advantages for global optimization suggest that quantum computers may be useful for training existing models within machine learning more quickly, and we are building experimental quantum computers to investigate how intricate quantum systems can carry out these computations. While this may prove invaluable, it does not yet touch on the tantalizing idea that quantum computers might be able to provide a way to learn more about complex patterns in physical systems that conventional computers cannot in any reasonable amount of time.

Today we talk about two recent papers from the Google AI Quantum team that make progress towards understanding the power of quantum computers for learning tasks. The first constructs a quantum model of neural networks to investigate how a popular classification task might be carried out on quantum processors. In the second paper, we show how peculiar features of quantum geometry change the strategies for training these networks in comparison to their classical counterparts, and offer guidance towards more robust training of these networks.

In “Classification with Quantum Neural Networks on Near Term Processors”, we construct a model of quantum neural networks (QNNs) that is specifically designed to work on quantum processors that are expected to be available in the near term. While the current work is primarily theoretical, their structure facilitates implementation and testing on quantum computers in the immediate future. These QNNs can be adapted through supervised learning of labeled data, and we show that it is possible to train a QNN to classify images in the famous MNIST dataset. Follow up work in this area with larger quantum devices may pit the ability of quantum networks to learn patterns against popular classical networks.
Quantum Neural Network for classification. Here we depict a sample quantum neural network, where in contrast to hidden layers in classical deep neural networks, the boxes represent entangling actions, or “quantum gates”, on qubits. In a superconducting qubit setup this could be enacted through a microwave control pulse corresponding to each box.
In “Barren Plateaus in Quantum Neural Network Training Landscapes”, we focus on the training of quantum neural networks, and probe questions related to a key difficulty in classical neural networks, which is the problem of vanishing or exploding gradients. In conventional neural networks, a good unbiased initial guess for the neuron weights often involves randomization, although there can be some difficulties as well. Our paper shows that peculiar features of quantum geometry unequivocally prevent this from being a good strategy in the quantum case, instead taking you to barren plateaus. The implications of this work may guide future strategies for initializing and training quantum neural networks.
QNN vanishing gradient: concentration of measure in high dimensional spaces. In very high dimensional spaces, such as those explored by quantum computers, the vast majority of states counterintuitively sit near the equator of the hypersphere (left). This means that any smooth function on this space will tend to take a value very close to its mean with overwhelming probability when selected at random (right).
This research sets the stage for improvements in both the construction and training of quantum neural networks. In particular, experimental realizations of quantum neural networks using hardware at Google will enable rapid exploration of quantum neural networks in the near term. We hope that the insights from the geometry of these states will lead to new algorithms to train these networks that will be essential to unlocking their full potential.

Source: Google AI Blog


Grasp2Vec: Learning Object Representations from Self-Supervised Grasping



From a remarkably young age, people are capable of recognizing their favorite objects and picking them up, despite never being explicitly taught how to do so. According to cognitive developmental research, the ability to interact with objects in the world plays a crucial role in the emergence of object perception and manipulation capabilities, such as targeted grasping. By interacting with the world around them, people are able to learn with self-supervision: we know what actions we took, and we learn from the outcome. In robotics, this type of self-supervised learning is actively researched because it enables robotic systems to learn without the need for large amounts of training data or manual supervision.

Inspired by the concept of object permanence, we propose Grasp2Vec, a simple yet highly effective algorithm for acquiring object representations. Grasp2Vec is based on the intuition that an attempt to pick up anything provides several pieces of information — if a robot grasps an object and holds it up, the object had to be in the scene before the grasp. Furthermore, the robot knows that the object it grasped is currently in its gripper, and therefore has been removed from the scene. By using this form of self supervision, the robot can learn to recognize the object by the visual change in the scene after the grasp.
Building on our prior collaboration with X Robotics, where a series of robots learn in parallel to grasp household objects using only monocular camera inputs, we use a robotic arm to grasp objects “unintentionally”, and that experience enables the learning of a rich representation of objects. These representations can then be used to acquire “intentional grasping” capabilities, where the robot arm can then pick up user-commanded objects.
Constructing a Perceptual Reward Function
In the framework of reinforcement learning (RL), task success is measured via a “reward function”. By maximizing that reward, robots can teach themselves diverse grasping skills from scratch. Engineering a reward function is easy when success can be measured by simple sensor measurements. A simple example of this is a button that supplies rewards directly to a robot when it is pushed.

However, engineering a reward function is much more difficult when our success criteria depends on perceptual understanding of the task at hand. Consider the task of instance grasping, where a robot is presented a picture of a desired object being held in the gripper. After the robot attempts to grasp that object, it inspects the contents of the gripper. The reward function for this task comes down to answering the question of object recognition: Do these objects match?
On the left, the gripper is holding the brush and there are some objects (yellow cup, blue plastic block) in the background. On the right, the gripper is holding the yellow cup and the brush is in the background. If the left image was the desired outcome, a good reward function should “understand” that the two images above correspond to different objects.
In order to solve this recognition problem, we need a perception system that extracts meaningful object concepts from unstructured image data (without any human annotations), learning the visual perception of objects in an unsupervised fashion. At their core, unsupervised learning algorithms work because they make structural assumptions about data. It is common to assume that images can be compressed into a low-dimensional space, and that frames in a video can be predicted from previous frames. However, without further assumptions on the content of the data, these are usually insufficient for learning disentangled object representations.

What if we used a robot to physically disentangle objects from each other during data collection? The field of robotics presents an exciting opportunity for representation learning because robots can manipulate objects, thus providing the factors of variation needed in data. Our method relies on the insight that grasping an object removes it from the scene. This yields 1) an image of the scene before grasping, 2) an image of the scene after grasping and 3) an isolated view of the grasped object itself.
Left: Objects before the grasp. Center: Objects after the grasp. Right: The Grasped object.
If we then consider an embedding function that extracts “the set of objects” from images, it should preserve the following subtractive relation:
objects_before_grasp - objects_after_grasp = grasped_object
We implement this equality relation using a fully convolutional architecture and a simple metric learning algorithm. At training time, the architecture shown below embeds the pre-grasp images and post-grasp images into a dense spatial feature map. The maps are mean-pooled into vectors and the difference between the “before grasp” and “after grasp” vectors represents a set of objects. This vector and the corresponding vector representation of the grasped object are pushed to equivalence via the N-Pairs objective.
Add caption
Once trained, two useful properties emerge naturally from our model.

1. Object Similarity
The first property is that a cosine distance between vector embeddings allows us to compare objects and determine whether they are identical. This can be used to implement reward functions for reinforcement learning, and allow robots to learn instance grasping without human-provided labels.
2. Localizing Target Objects
The second property is that we can combine scene spatial maps and object embeddings to localize a “query object” in image space. By taking the element-wise product of spatial feature maps and the vector corresponding to the query object, we can find all the pixels in the spatial map that “match” the query object.
Using Grasp2Vec embeddings to localize objects in a scene. The image on the top left shows the objects in the bin. On the bottom left is the query object we wish to grasp. By taking the dot product of the query object vector with the spatial features of the scene image, we get a per-pixel “activation map” (top right image) of how similar that region of the image is to the query. This response map can be used to approach the object for grasping.
Our method also works when there are multiple objects that match the query object, or even if the query consists of multiple objects (the average of two vectors). For example, here is a scenario where it detects multiple orange blocks in a scene.
The resulting “heatmap” can be used to plan the robot approach to the target object(s). We combine Grasp2Vec’s localization and instance recognition capabilities with our “grasp anything” policies to obtain a success rate of 80% on objects seen during data collection and 59% on novel objects the robot hasn’t encountered before.

Conclusion
In our paper, we show how robotic grasping skills can generate the data used for learning object-centric representations. We then can use representation learning to “bootstrap” more complex skills like instance grasping, all while retaining the self-supervised learning properties of our autonomous grasping system.

Besides our own work, a number of recent papers have also studied how self-supervised interaction can be used to acquire representations, by grasping, pushing, and otherwise manipulating objects in the environment. Going forward, we are excited not only for what machine learning can bring to robotics by way of better perception and control, but also what robotics can bring to machine learning in new paradigms of self-supervision.

Acknowledgements
This research was conducted by Eric Jang, Coline Devin, Vincent Vanhoucke, and Sergey Levine. We’d like to thank Adrian Li, Alex Irpan, Anthony Brohan, Chelsea Finn, Christian Howard, Corey Lynch, Dmitry Kalashnikov, Ian Wilkes, Ivonne Fajardo, Julian Ibarz, Ming Zhao, Peter Pastor, Pierre Sermanet, Stephen James, Tsung-Yi Lin, Yunfei Bai, and many others at Google, X, and the broader robotics community who contributed to improving this work.

Source: Google AI Blog


Google at NeurIPS 2018



This week, Montréal hosts the 32nd annual Conference on Neural Information Processing Systems (NeurIPS 2018), the biggest machine learning conference of the year. The conference includes invited talks, demonstrations and presentations of some of the latest in machine learning research. Google will have a strong presence at NeurIPS 2018, with more than 400 Googlers attending in order to contribute to, and learn from, the broader academic research community via talks, posters, workshops, competitions and tutorials. We will be presenting work that pushes the boundaries of what is possible in language understanding, translation, speech recognition and visual & audio perception, with Googlers co-authoring nearly 100 accepted papers (see below).

At the forefront of machine learning, Google is actively exploring virtually all aspects of the field spanning both theory and applications. This research is often inspired by real product needs but increasingly more often driven by scientific curiosity. Given the range of research projects that we pursue, we have found it useful to define a new framework which helps crystalize the goals of projects and allows us to measure progress and success in appropriate ways. Our contributions to NeurIPS and to the broader research community in general are integral to our research mission.

If you are attending NeurIPS 2018, we hope you’ll stop by our booth and chat with our researchers about the projects and opportunities at Google that go into solving the world's most challenging research problems, and to see demonstrations of some of the exciting research we pursue. You can also learn more about our work being presented in the list below (Googlers highlighted in blue).

Google is a Platinum Sponsor of NeurIPS 2018.

NeurIPS Foundation Board
Corinna Cortes, John C. Platt, Fernando Pereira

NeurIPS Organizing Committee
General Chair: Samy Bengio
Program Co-Chair: Hugo Larochelle
Party Chair: Douglas Eck
Diversity and Inclusion Co-Chair: Katherine A. Heller

NeurIPS Program Committee
Senior Area Chairs include:Angela Yu, Claudio Gentile, Cordelia Schmid, Corinna Cortes, Csaba Szepesvari, Dale Schuurmans, Elad Hazan, Mehryar Mohri, Raia Hadsell, Satyen Kale, Yishay Mansour, Afshin Rostamizadeh, Alex Kulesza

Area Chairs include: Amin Karbasi, Amir Globerson, Amit Daniely, Andras Gyorgy, Andriy Mnih, Been Kim, Branislav Kveton, Ce Liu, D Sculley, Danilo Rezende, Danny TarlowDavid Balduzzi, Denny Zhou, Dilan Gorur, Dumitru Erhan, George Dahl, Graham Taylor, Ian Goodfellow, Jasper Snoek, Jean-Philippe Vert, Jia Deng, Jon Shlens, Karen Simonyan, Kevin Swersky, Kun Zhang, Lihong Li, Marc G. Bellemare, Marco Cuturi, Maya Gupta, Michael BowlingMichalis Titsias, Mohammad Norouzi, Mouhamadou Moustapha Cisse, Nicolas Le Roux, Remi Munos, Sanjiv Kumar, Sanmi Koyejo, Sergey Levine, Silvia Chiappa, Slav PetrovSurya Ganguli, Timnit Gebru, Timothy Lillicrap, Viren Jain, Vitaly Feldman, Vitaly Kuznetsov

Workshops Program Committee includes: Mehryar Mohri, Sergey Levine

Accepted Papers
3D-Aware Scene Manipulation via Inverse Graphics
Shunyu Yao, Tzu Ming Harry Hsu, Jun-Yan Zhu, Jiajun Wu, Antonio Torralba, William T. Freeman, Joshua B. Tenenbaum

A Retrieve-and-Edit Framework for Predicting Structured Outputs
Tatsunori Hashimoto, Kelvin Guu, Yonatan Oren, Percy Liang

Adversarial Attacks on Stochastic Bandits
Kwang-Sung Jun, Lihong Li, Yuzhe Ma, Xiaojin Zhu

Adversarial Examples that Fool both Computer Vision and Time-Limited Humans
Gamaleldin F. Elsayed, Shreya Shankar, Brian Cheung, Nicolas Papernot, Alex Kurakin, Ian Goodfellow, Jascha Sohl-Dickstein

Adversarially Robust Generalization Requires More Data
Ludwig Schmidt, Shibani Santurkar, Dimitris Tsipras, Kunal Talwar, Aleksander Madry

Are GANs Created Equal? A Large-Scale Study
Mario Lucic, Karol Kurach, Marcin Michalski, Olivier Bousquet, Sylvain Gelly

Collaborative Learning for Deep Neural Networks
Guocong Song, Wei Chai

Completing State Representations using Spectral Learning
Nan Jiang, Alex Kulesza, Santinder Singh

Content Preserving Text Generation with Attribute Controls
Lajanugen Logeswaran, Honglak Lee, Samy Bengio

Context-aware Synthesis and Placement of Object Instances
Donghoon Lee, Sifei Liu, Jinwei Gu, Ming-Yu Liu, Ming-Hsuan Yang, Jan Kautz

Co-regularized Alignment for Unsupervised Domain Adaptation
Abhishek Kumar, Prasanna Sattigeri, Kahini Wadhawan, Leonid Karlinsky, Rogerlo Feris, William T. Freeman, Gregory Wornell

cpSGD: Communication-efficient and differentially-private distributed SGD
Naman Agarwal, Ananda Theertha Suresh, Felix Yu, Sanjiv Kumar, H. Brendan Mcmahan

Data Center Cooling Using Model-Predictive Control
Nevena Lazic, Craig Boutilier, Tyler Lu, Eehern Wong, Binz Roy, MK Ryu, Greg Imwalle

Data-Efficient Hierarchical Reinforcement Learning
Ofir Nachum, Shixiang Gu, Honglak Lee, Sergey Levine

Deep Attentive Tracking via Reciprocative Learning
Shi Pu, Yibing Song, Chao Ma, Honggang Zhang, Ming-Hsuan Yang

Generalizing Point Embeddings Using the Wasserstein Space of Elliptical Distributions
Boris Muzellec, Marco Cuturi

GLoMo: Unsupervised Learning of Transferable Relational Graphs
Zhilin Yang, Jake (Junbo) Zhao, Bhuwan Dhingra, Kaiming He, William W. Cohen, Ruslan Salakhutdinov, Yann LeCun

GroupReduce: Block-Wise Low-Rank Approximation for Neural Language Model Shrinking
Patrick Chen, Si Si, Yang Li, Ciprian Chelba, Cho-Jui Hsieh

Interpreting Neural Network Judgments via Minimal, Stable, and Symbolic Corrections
Xin Zhang, Armando Solar-Lezama, Rishabh Singh

Learning Hierarchical Semantic Image Manipulation through Structured Representations
Seunghoon Hong, Xinchen Yan, Thomas Huang, Honglak Lee

Learning Temporal Point Processes via Reinforcement Learning
Shuang Li, Shuai Xiao, Shixiang Zhu, Nan Du, Yao Xie, Le Song

Learning Towards Minimum Hyperspherical Energy
Weiyang Liu, Rongmei Lin, Zhen Liu, Lixin Liu, Zhiding Yu, Bo Dai, Le Song

Mesh-TensorFlow: Deep Learning for Supercomputers
Noam Shazeer, Youlong Cheng, Niki Parmar, Dustin Tran, Ashish Vaswani, Penporn Koanantakool, Peter Hawkins, HyoukJoong Lee, Mingsheng Hong, Cliff Young, Ryan Sepassi, Blake Hechtman

MiME: Multilevel Medical Embedding of Electronic Health Records for Predictive Healthcare
Edward Choi, Cao Xiao, Walter F. Stewart, Jimeng Sun

Searching for Efficient Multi-Scale Architectures for Dense Image Prediction
Liang-Chieh Chen, Maxwell D. Collins, Yukun Zhu, George Papandreou, Barret Zoph, Florian Schroff, Hartwig Adam, Jonathon Shlens

SplineNets: Continuous Neural Decision Graphs
Cem Keskin, Shahram Izadi

Task-Driven Convolutional Recurrent Models of the Visual System
Aran Nayebi, Daniel Bear, Jonas Kubilius, Kohitij Kar, Surya Ganguli, David Sussillo, James J. DiCarlo, Daniel L. K. Yamins

To Trust or Not to Trust a Classifier
Heinrich Jiang, Been Kim, Melody Guan, Maya Gupta

Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis
Ye Jia, Yu Zhang, Ron J. Weiss, Quan Wang, Jonathan Shen, Fei Ren, Zhifeng Chen, Patrick Nguyen, Ruoming Pang, Ignacio Lopez Moreno, Yonghui Wu

Algorithms and Theory for Multiple-Source Adaptation
Judy Hoffman, Mehryar Mohri, Ningshan Zhang

A Lyapunov-based Approach to Safe Reinforcement Learning
Yinlam Chow, Ofir Nachum, Edgar Duenez-Guzman, Mohammad Ghavamzadeh

Adaptive Methods for Nonconvex Optimization
Manzil Zaheer, Sashank Reddi, Devendra Sachan, Satyen Kale, Sanjiv Kumar

Assessing Generative Models via Precision and Recall
Mehdi S. M. Sajjadi, Olivier Bachem, Mario Lucic, Olivier Bousquet, Sylvain Gelly

A Loss Framework for Calibrated Anomaly Detection
Aditya Menon, Robert Williamson

Blockwise Parallel Decoding for Deep Autoregressive Models
Mitchell Stern, Noam Shazeer, Jakob Uszkoreit

Breaking the Curse of Horizon: Infinite-Horizon Off-Policy Estimation
Qiang Liu, Lihong Li, Ziyang Tang, Dengyong Zhou

Contextual Pricing for Lipschitz Buyers
Jieming Mao, Renato Leme, Jon Schneider

Coupled Variational Bayes via Optimization Embedding
Bo Dai, Hanjun Dai, Niao He, Weiyang Liu, Zhen Liu, Jianshu Chen, Lin Xiao, Le Song

Data Amplification: A Unified and Competitive Approach to Property Estimation
Yi HAO, Alon Orlitsky, Ananda Theertha Suresh, Yihong Wu

Deep Network for the Integrated 3D Sensing of Multiple People in Natural Images
Elisabeta Marinoiu, Mihai Zanfir, Alin-Ionut Popa, Cristian Sminchisescu

Deep Non-Blind Deconvolution via Generalized Low-Rank Approximation
Wenqi Ren, Jiawei Zhang, Lin Ma, Jinshan Pan, Xiaochun Cao, Wei Liu, Ming-Hsuan Yang

Diminishing Returns Shape Constraints for Interpretability and Regularization
Maya Gupta, Dara Bahri, Andrew Cotter, Kevin Canini

DropBlock: A Regularization Method for Convolutional Networks
Golnaz Ghiasi, Tsung-Yi Lin, Quoc V. Le

Generalization Bounds for Uniformly Stable Algorithms
Vitaly Feldman, Jan Vondrak

Geometrically Coupled Monte Carlo Sampling
Mark Rowland, Krzysztof Choromanski, Francois Chalus, Aldo Pacchiano, Tamas Sarlos, Richard E. Turner, Adrian Weller

GILBO: One Metric to Measure Them All
Alexander A. Alemi, Ian Fischer

Insights on Representational Similarity in Neural Networks with Canonical Correlation
Ari S. Morcos, Maithra Raghu, Samy Bengio

Improving Online Algorithms via ML Predictions
Manish Purohit, Zoya Svitkina, Ravi Kumar

Learning to Exploit Stability for 3D Scene Parsing
Yilun Du, Zhijan Liu, Hector Basevi, Ales Leonardis, William T. Freeman, Josh Tenembaum, Jiajun Wu

Maximizing Induced Cardinality Under a Determinantal Point Process
Jennifer Gillenwater, Alex Kulesza, Sergei Vassilvitskii, Zelda Mariet

Memory Augmented Policy Optimization for Program Synthesis and Semantic Parsing
Chen Liang, Mohammad Norouzi, Jonathan Berant, Quoc V. Le, Ni Lao

PCA of High Dimensional Random Walks with Comparison to Neural Network Training
Joseph M. Antognini, Jascha Sohl-Dickstein

Predictive Approximate Bayesian Computation via Saddle Points
Yingxiang Yang, Bo Dai, Negar Kiyavash, Niao He

Recurrent World Models Facilitate Policy Evolution
David Ha, Jürgen Schmidhuber

Sanity Checks for Saliency Maps
Julius Adebayo, Justin Gilmer, Michael Muelly, Ian Goodfellow, Moritz Hardt, Been Kim

Simple, Distributed, and Accelerated Probabilistic Programming
Dustin Tran, Matthew Hoffman, Dave Moore, Christopher Suter, Srinivas Vasudevan, Alexey Radul, Matthew Johnson, Rif A. Saurous

Tangent: Automatic Differentiation Using Source-Code Transformation for Dynamically Typed Array Programming
Bart van Merriënboer, Dan Moldovan, Alex Wiltschko

The Emergence of Multiple Retinal Cell Types Through Efficient Coding of Natural Movies
Samuel A. Ocko, Jack Lindsey, Surya Ganguli, Stephane Deny

The Everlasting Database: Statistical Validity at a Fair Price
Blake Woodworth, Vitaly Feldman, Saharon Rosset, Nathan Srebro

The Spectrum of the Fisher Information Matrix of a Single-Hidden-Layer Neural Network
Jeffrey Pennington, Pratik Worah

A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks
Kimin Lee, Kibok Lee, Honglak Lee, Jinwoo Shin

Autoconj: Recognizing and Exploiting Conjugacy Without a Domain-Specific Language
Matthew D. Hoffman, Matthew Johnson, Dustin Tran

A Bayesian Nonparametric View on Count-Min Sketch
Diana Cai, Michael Mitzenmacher, Ryan Adams (no longer at Google)

Automatic Differentiation in ML: Where We are and Where We Should be Going
Bart van Merriënboer, Olivier Breuleux, Arnaud Bergeron, Pascal Lamblin

Assessing the Scalability of Biologically-Motivated Deep Learning Algorithms and Architectures
Sergey Bartunov, Adam Santoro, Blake A. Richards, Geoffrey E. Hinton, Timothy P. Lillicrap

Deep Generative Models for Distribution-Preserving Lossy Compression
Michael Tschannen, Eirikur Agustsson, Mario Lucic

Deep Structured Prediction with Nonlinear Output Transformations
Colin Graber, Ofer Meshi, Alexander Schwing

Discovery of Latent 3D Keypoints via End-to-end Geometric Reasoning
Supasorn Suwajanakorn, Noah Snavely, Jonathan Tompson, Mohammad Norouzi

Transfer Learning with Neural AutoML
Catherine Wong, Neil Houlsby, Yifeng Lu, Andrea Gesmundo

Efficient Gradient Computation for Structured Output Learning with Rational and Tropical Losses
Corinna Cortes, Vitaly Kuznetsov, Mehryar Mohri, Dmitry Storcheus, Scott Yang

Cooperative neural networks (CoNN): Exploiting prior independence structure for improved classification
Harsh Shrivastava, Eugene Bart, Bob Price, Hanjun Dai, Bo Dai, Srinivas Aluru

Graph Oracle Models, Lower Bounds, and Gaps for Parallel Stochastic Optimization
Blake Woodworth, Jialei Wang, Brendan McMahan, Nathan Srebro

Hierarchical Reinforcement Learning for Zero-shot Generalization with Subtask Dependencies
Sungryull Sohn, Junhyuk Oh, Honglak Lee

Human-in-the-Loop Interpretability Prior
Isaac Lage, Andrew Slavin Ross, Been Kim, Samuel J. Gershman, Finale Doshi-Velez

Joint Autoregressive and Hierarchical Priors for Learned Image Compression
David Minnen, Johannes Ballé, George D Toderici

Large-Scale Computation of Means and Clusters for Persistence Diagrams Using Optimal Transport
Théo Lacombe, Steve Oudot, Marco Cuturi

Learning to Reconstruct Shapes from Unseen Classes
Xiuming Zhang, Zhoutong Zhang, Chengkai Zhang, Joshua B. Tenenbaum, William T. Freeman, Jiajun Wu

Large Margin Deep Networks for Classification
Gamaleldin Fathy Elsayed, Dilip Krishnan, Hossein Mobahi, Kevin Regan, Samy Bengio

Mallows Models for Top-k Lists
Flavio Chierichetti, Anirban Dasgupta, Shahrzad Haddadan, Ravi Kumar, Silvio Lattanzi

Meta-Learning MCMC Proposals
Tongzhou Wang, YI WU, Dave Moore, Stuart Russell

Non-delusional Q-Learning and Value-Iteration
Tyler Lu, Dale Schuurmans, Craig Boutilier

Online Learning of Quantum States
Scott Aaronson, Xinyi Chen, Elad Hazan, Satyen Kale, Ashwin Nayak

Online Reciprocal Recommendation with Theoretical Performance Guarantees
Fabio Vitale, Nikos Parotsidis, Claudio Gentile

Optimal Algorithms for Continuous Non-monotone Submodular and DR-Submodular Maximization
Rad Niazadeh, Tim Roughgarden, Joshua R. Wang

Policy Regret in Repeated Games
Raman Arora, Michael Dinitz, Teodor Vanislavov Marinov, Mehryar Mohri

Provable Variational Inference for Constrained Log-Submodular Models
Josip Djolonga, Stefanie Jegelka, Andreas Krause

Realistic Evaluation of Deep Semi-Supervised Learning Algorithms
Avital Oliver, Augustus Odena, Colin Raffel, Ekin D. Cubuk, Ian J. Goodfellow

Sample-Efficient Reinforcement Learning with Stochastic Ensemble Value Expansion
Jacob Buckman, Danijar Hafner, George Tucker, Eugene Brevdo, Honglak Lee

Visual Object Networks: Image Generation with Disentangled 3D Representations
JunYan Zhu, Zhoutong Zhang, Chengkai Zhang, Jiajun Wu, Antonio Torralba, Josh Tenenbaum, William T. Freeman

Watch Your Step: Learning Node Embeddings via Graph Attention
Sami Abu-El-Haija, Bryan Perozzi, Rami AlRfou, Alexander Alemi

Workshops
2nd Workshop on Machine Learning on the Phone and Other Consumer Devices
Co-Chairs include: Sujith Ravi, Wei Chai, Hrishikesh Aradhye

Bayesian Deep Learning
Workshop Organizers include: Kevin Murphy

Continual Learning
Workshop Organizers include: Marc Pickett

The Second Conversational AI Workshop – Today's Practice and Tomorrow's Potential
Workshop Organizers include: Dilek Hakkani-Tur

Visually Grounded Interaction and Language
Workshop Organizers include: Olivier Pietquin

Workshop on Ethical, Social and Governance Issues in AI
Workshop Organizers include: D. Sculley

AI for Social Good
Workshop Program Committee includes: Samuel Greydanus

Black in AI
Workshop Organizers: Mouhamadou Moustapha Cisse, Timnit Gebru
Program Committee: Irwan Bello, Samy Bengio, Ian Goodfellow, Hugo Larochelle, Margaret Mitchell

Interpretability and Robustness in Audio, Speech, and Language
Workshop Organizers include: Ehsan Variani, Bhuvana Ramabhadran

LatinX in AI
Workshop Organizers includes: Pablo Samuel Castro
Program Committee includes: Sergio Guadarrama

Machine Learning for Systems
Workshop Organizers include: Anna Goldie, Azalia Mirhoseini, Kevin Swersky, Milad Hashemi
Program Committee includes: Simon Kornblith, Nicholas Frosst, Amir Yazdanbakhsh, Azade Nazi, James Bradbury, Sharan Narang, Martin Maas, Carlos Villavieja

Queer in AI
Workshop Organizers include: Raphael Gontijo Lopes

Second Workshop on Machine Learning for Creativity and Design
Workshop Organizers include: Jesse Engel, Adam Roberts

Workshop on Security in Machine Learning
Workshop Organizers include: Nicolas Papernot

Tutorial
Visualization for Machine Learning
Fernanda Viégas, Martin Wattenberg

Source: Google AI Blog


Learning to Predict Depth on the Pixel 3 Phones



Portrait Mode on the Pixel smartphones lets you take professional-looking images that draw attention to a subject by blurring the background behind it. Last year, we described, among other things, how we compute depth with a single camera using its Phase-Detection Autofocus (PDAF) pixels (also known as dual-pixel autofocus) using a traditional non-learned stereo algorithm. This year, on the Pixel 3, we turn to machine learning to improve depth estimation to produce even better Portrait Mode results.
Left: The original HDR+ image. Right: A comparison of Portrait Mode results using depth from traditional stereo and depth from machine learning. The learned depth result has fewer errors. Notably, in the traditional stereo result, many of the horizontal lines behind the man are incorrectly estimated to be at the same depth as the man and are kept sharp.
(Mike Milne)
A Short Recap
As described in last year’s blog post, Portrait Mode uses a neural network to determine what pixels correspond to people versus the background, and augments this two layer person segmentation mask with depth information derived from the PDAF pixels. This is meant to enable a depth-dependent blur, which is closer to what a professional camera does.

PDAF pixels work by capturing two slightly different views of a scene, shown below. Flipping between the two views, we see that the person is stationary, while the background moves horizontally, an effect referred to as parallax. Because parallax is a function of the point’s distance from the camera and the distance between the two viewpoints, we can estimate depth by matching each point in one view with its corresponding point in the other view.
The two PDAF images on the left and center look very similar, but in the crop on the right you can see the parallax between them. It is most noticeable on the circular structure in the middle of the crop.
However, finding these correspondences in PDAF images (a method called depth from stereo) is extremely challenging because scene points barely move between the views. Furthermore, all stereo techniques suffer from the aperture problem. That is, if you look at the scene through a small aperture, it is impossible to find correspondence for lines parallel to the stereo baseline, i.e., the line connecting the two cameras. In other words, when looking at the horizontal lines in the figure above (or vertical lines in portrait orientation shots), any proposed shift of these lines in one view with respect to the other view looks about the same. In last year’s Portrait Mode, all these factors could result in errors in depth estimation and cause unpleasant artifacts.

Improving Depth Estimation
With Portrait Mode on the Pixel 3, we fix these errors by utilizing the fact that the parallax used by depth from stereo algorithms is only one of many depth cues present in images. For example, points that are far away from the in-focus plane appear less sharp than ones that are closer, giving us a defocus depth cue. In addition, even when viewing an image on a flat screen, we can accurately tell how far things are because we know the rough size of everyday objects (e.g. one can use the number of pixels in a photograph of a person’s face to estimate how far away it is). This is called a semantic cue.

Designing a hand-crafted algorithm to combine these different cues is extremely difficult, but by using machine learning, we can do so while also better exploiting the PDAF parallax cue. Specifically, we train a convolutional neural network, written in TensorFlow, that takes as input the PDAF pixels and learns to predict depth. This new and improved ML-based method of depth estimation is what powers Portrait Mode on the Pixel 3.
Our convolutional neural network takes as input the PDAF images and outputs a depth map. The network uses an encoder-decoder style architecture with skip connections and residual blocks.
Training the Neural Network
In order to train the network, we need lots of PDAF images and corresponding high-quality depth maps. And since we want our predicted depth to be useful for Portrait Mode, we also need the training data to be similar to pictures that users take with their smartphones.

To accomplish this, we built our own custom “Frankenphone” rig that contains five Pixel 3 phones, along with a Wi-Fi-based solution that allowed us to simultaneously capture pictures from all of the phones (within a tolerance of ~2 milliseconds). With this rig, we computed high-quality depth from photos by using structure from motion and multi-view stereo.
Left: Custom rig used to collect training data. Middle: An example capture flipping between the five images. Synchronization between the cameras ensures that we can calculate depth for dynamic scenes, such as this one. Right: Ground truth depth. Low confidence points, i.e., points where stereo matches are not reliable due to weak texture, are colored in black and are not used during training. (Sam Ansari and Mike Milne)
The data captured by this rig is ideal for training a network for the following main reasons:
  • Five viewpoints ensure that there is parallax in multiple directions and hence no aperture problem.
  • The arrangement of the cameras ensures that a point in an image is usually visible in at least one other image resulting in fewer points with no correspondences.
  • The baseline, i.e., the distance between the cameras is much larger than our PDAF baseline resulting in more accurate depth estimation.
  • Synchronization between the cameras ensure that we can calculate depth for dynamic scenes like the one above.
  • Portability of the rig ensures that we can capture photos in the wild simulating the photos users take with their smartphones.
However, even though the data captured from this rig is ideal, it is still extremely challenging to predict the absolute depth of objects in a scene — a given PDAF pair can correspond to a range of different depth maps (depending on lens characteristics, focus distance, etc). To account for this, we instead predict the relative depths of objects in the scene, which is sufficient for producing pleasing Portrait Mode results.

Putting it All Together
This ML-based depth estimation needs to run fast on the Pixel 3, so that users don’t have to wait too long for their Portrait Mode shots. However, to get good depth estimates that makes use of subtle defocus and parallax cues, we have to feed full resolution, multi-megapixel PDAF images into the network. To ensure fast results, we use TensorFlow Lite, a cross-platform solution for running machine learning models on mobile and embedded devices and the Pixel 3’s powerful GPU to compute depth quickly despite our abnormally large inputs. We then combine the resulting depth estimates with masks from our person segmentation neural network to produce beautiful Portrait Mode results.

Try it Yourself
In Google Camera App version 6.1 and later, our depth maps are embedded in Portrait Mode images. This means you can use the Google Photos depth editor to change the amount of blur and the focus point after capture. You can also use third-party depth extractors to extract the depth map from a jpeg and take a look at it yourself. Also, here is an album showing the relative depth maps and the corresponding Portrait Mode images for traditional stereo and the learning-based approaches.

Acknowledgments
This work wouldn’t have been possible without Sam Ansari, Yael Pritch Knaan, David Jacobs, Jiawen Chen, Juhyun Lee and Andrei Kulik. Special thanks to Mike Milne and Andy Radin who captured data with the five-camera rig.

Source: Google AI Blog


Introduction to Fairness in Machine Learning

Posted by Andrew Zaldivar, Developer Advocate, Google AI

A few months ago, we announced our AI Principles, a set of commitments we are upholding to guide our work in artificial intelligence (AI) going forward. Along with our AI Principles, we shared a set of recommended practices to help the larger community design and build responsible AI systems.

In particular, one of our AI Principles speaks to the importance of recognizing that AI algorithms and datasets are the product of the environment—and, as such, we need to be conscious of any potential unfair outcomes generated by an AI system and the risk it poses across cultures and societies. A recommended practice here for practitioners is to understand the limitations of their algorithm and datasets—but this is a problem that is far from solved.

To help practitioners take on the challenge of building fairer and more inclusive AI systems, we developed a short, self-study training module on fairness in machine learning. This new module is part of our Machine Learning Crash Course, which we highly recommend taking first—unless you know machine learning really well, in which case you can jump right into the Fairness module.

The Fairness module features a hands-on technical exercise. This exercise demonstrates how you can use tools and techniques that may already exist in your development stack (such as Facets Dive, Seaborn, pandas, scikit-learn and TensorFlow Estimators to name a few) to explore and discover ways to make your machine learning system fairer and more inclusive. We created our exercise in a Colaboratory notebook, which you are more than welcome to use, modify and distribute for your own purposes.

From exploring datasets to analyzing model performance, it's really easy to forget to make time for responsible reflection when building an AI system. So rather than having you run every code cell in sequential order without pause, we added what we call FairAware tasks throughout the exercise. FairAware tasks help you zoom in and out of the problem space. That way, you can remind yourself of the big picture: finding the undesirable biases that could disproportionately affect model performance across groups. We hope a process like FairAware will become part of your workflow, helping you find opportunities for inclusion.

FairAware task guiding practitioner to compare performances across gender.

The Fairness module was created to provide you with enough of an understanding to get started in addressing fairness and inclusion in AI. Keep an eye on this space for future work as this is only the beginning.

If you wish to learn more from our other examples, check out the Fairness section of our Responsible AI Practices guide. There, you will find a full set of Google recommendations and resources. From our latest research proposal on reporting model performance with fairness and inclusion considerations, to our recently launched diagnostic tool that lets anyone investigate trained models for fairness, our resource guide highlights many areas of research and development in fairness.

Let us know what your thoughts are on our Fairness module. If you have any specific comments on the notebook exercise itself, then feel free to leave a comment on our GitHub repo.


On behalf of many contributors and supporters,

Andrew Zaldivar – Developer Advocate, Google AI

Combating Potentially Harmful Applications with Machine Learning at Google: Datasets and Models

Posted by Mo Yu, Android Security & Privacy Team

In a previous blog post, we talked about using machine learning to combat Potentially Harmful Applications (PHAs). This blog post covers how Google uses machine learning techniques to detect and classify PHAs. We'll discuss the challenges in the PHA detection space, including the scale of data, the correct identification of PHA behaviors, and the evolution of PHA families. Next, we will introduce two of the datasets that make the training and implementation of machine learning models possible, such as app analysis data and Google Play data. Finally, we will present some of the approaches we use, including logistic regression and deep neural networks.

Using machine learning to scale

Detecting PHAs is challenging and requires a lot of resources. Our security experts need to understand how apps interact with the system and the user, analyze complex signals to find PHA behavior, and evolve their tactics to stay ahead of PHA authors. Every day, Google Play Protect (GPP) analyzes over half a million apps, which makes a lot of new data for our security experts to process.

Leveraging machine learning helps us detect PHAs faster and at a larger scale. We can detect more PHAs just by adding additional computing resources. In many cases, machine learning can find PHA signals in the training data without human intervention. Sometimes, those signals are different than signals found by security experts. Machine learning can take better advantage of this data, and discover hidden relationships between signals more effectively.

There are two major parts of Google Play Protect's machine learning protections: the data and the machine learning models.

Data sources

The quality and quantity of the data used to create a model are crucial to the success of the system. For the purpose of PHA detection and classification, our system mainly uses two anonymous data sources: data from analyzing apps and data from how users experience apps.

App data

Google Play Protect analyzes every app that it can find on the internet. We created a dataset by decomposing each app's APK and extracting PHA signals with deep analysis. We execute various processes on each app to find particular features and behaviors that are relevant to the PHA categories in scope (for example, SMS fraud, phishing, privilege escalation). Static analysis examines the different resources inside an APK file while dynamic analysis checks the behavior of the app when it's actually running. These two approaches complement each other. For example, dynamic analysis requires the execution of the app regardless of how obfuscated its code is (obfuscation hinders static analysis), and static analysis can help detect cloaking attempts in the code that may in practice bypass dynamic analysis-based detection. In the end, this analysis produces information about the app's characteristics, which serve as a fundamental data source for machine learning algorithms.

Google Play data

In addition to analyzing each app, we also try to understand how users perceive that app. User feedback (such as the number of installs, uninstalls, user ratings, and comments) collected from Google Play can help us identify problematic apps. Similarly, information about the developer (such as the certificates they use and their history of published apps) contribute valuable knowledge that can be used to identify PHAs. All these metrics are generated when developers submit a new app (or new version of an app) and by millions of Google Play users every day. This information helps us to understand the quality, behavior, and purpose of an app so that we can identify new PHA behaviors or identify similar apps.

In general, our data sources yield raw signals, which then need to be transformed into machine learning features for use by our algorithms. Some signals, such as the permissions that an app requests, have a clear semantic meaning and can be directly used. In other cases, we need to engineer our data to make new, more powerful features. For example, we can aggregate the ratings of all apps that a particular developer owns, so we can calculate a rating per developer and use it to validate future apps. We also employ several techniques to focus in on interesting data.To create compact representations for sparse data, we use embedding. To help streamline the data to make it more useful to models, we use feature selection. Depending on the target, feature selection helps us keep the most relevant signals and remove irrelevant ones.

By combining our different datasets and investing in feature engineering and feature selection, we improve the quality of the data that can be fed to various types of machine learning models.

Models

Building a good machine learning model is like building a skyscraper: quality materials are important, but a great design is also essential. Like the materials in a skyscraper, good datasets and features are important to machine learning, but a great algorithm is essential to identify PHA behaviors effectively and efficiently.

We train models to identify PHAs that belong to a specific category, such as SMS-fraud or phishing. Such categories are quite broad and contain a large number of samples given the number of PHA families that fit the definition. Alternatively, we also have models focusing on a much smaller scale, such as a family, which is composed of a group of apps that are part of the same PHA campaign and that share similar source code and behaviors. On the one hand, having a single model to tackle an entire PHA category may be attractive in terms of simplicity but precision may be an issue as the model will have to generalize the behaviors of a large number of PHAs believed to have something in common. On the other hand, developing multiple PHA models may require additional engineering efforts, but may result in better precision at the cost of reduced scope.

We use a variety of modeling techniques to modify our machine learning approach, including supervised and unsupervised ones.

One supervised technique we use is logistic regression, which has been widely adopted in the industry. These models have a simple structure and can be trained quickly. Logistic regression models can be analyzed to understand the importance of the different PHA and app features they are built with, allowing us to improve our feature engineering process. After a few cycles of training, evaluation, and improvement, we can launch the best models in production and monitor their performance.

For more complex cases, we employ deep learning. Compared to logistic regression, deep learning is good at capturing complicated interactions between different features and extracting hidden patterns. The millions of apps in Google Play provide a rich dataset, which is advantageous to deep learning.

In addition to our targeted feature engineering efforts, we experiment with many aspects of deep neural networks. For example, a deep neural network can have multiple layers and each layer has several neurons to process signals. We can experiment with the number of layers and neurons per layer to change model behaviors.

We also adopt unsupervised machine learning methods. Many PHAs use similar abuse techniques and tricks, so they look almost identical to each other. An unsupervised approach helps define clusters of apps that look or behave similarly, which allows us to mitigate and identify PHAs more effectively. We can automate the process of categorizing that type of app if we are confident in the model or can request help from a human expert to validate what the model found.

PHAs are constantly evolving, so our models need constant updating and monitoring. In production, models are fed with data from recent apps, which help them stay relevant. However, new abuse techniques and behaviors need to be continuously detected and fed into our machine learning models to be able to catch new PHAs and stay on top of recent trends. This is a continuous cycle of model creation and updating that also requires tuning to ensure that the precision and coverage of the system as a whole matches our detection goals.

Looking forward

As part of Google's AI-first strategy, our work leverages many machine learning resources across the company, such as tools and infrastructures developed by Google Brain and Google Research. In 2017, our machine learning models successfully detected 60.3% of PHAs identified by Google Play Protect, covering over 2 billion Android devices. We continue to research and invest in machine learning to scale and simplify the detection of PHAs in the Android ecosystem.

Acknowledgements

This work was developed in joint collaboration with Google Play Protect, Safe Browsing and Play Abuse teams with contributions from Andrew Ahn, Hrishikesh Aradhye, Daniel Bali, Hongji Bao, Yajie Hu, Arthur Kaiser, Elena Kovakina, Salvador Mandujano, Melinda Miller, Rahul Mishra, Damien Octeau, Sebastian Porst, Chuangang Ren, Monirul Sharif, Sri Somanchi, Sai Deep Tetali, Zhikun Wang, and Mo Yu.

Introducing AdaNet: Fast and Flexible AutoML with Learning Guarantees



Ensemble learning, the art of combining different machine learning (ML) model predictions, is widely used with neural networks to achieve state-of-the-art performance, benefitting from a rich history and theoretical guarantees to enable success at challenges such as the Netflix Prize and various Kaggle competitions. However, they aren’t used much in practice due to long training times, and the ML model candidate selection requires its own domain expertise. But as computational power and specialized deep learning hardware such as TPUs become more readily available, machine learning models will grow larger and ensembles will become more prominent. Now, imagine a tool that automatically searches over neural architectures, and learns to combine the best ones into a high-quality model.

Today, we’re excited to share AdaNet, a lightweight TensorFlow-based framework for automatically learning high-quality models with minimal expert intervention. AdaNet builds on our recent reinforcement learning and evolutionary-based AutoML efforts to be fast and flexible while providing learning guarantees. Importantly, AdaNet provides a general framework for not only learning a neural network architecture, but also for learning to ensemble to obtain even better models.

AdaNet is easy to use, and creates high-quality models, saving ML practitioners the time normally spent selecting optimal neural network architectures, implementing an adaptive algorithm for learning a neural architecture as an ensemble of subnetworks. AdaNet is capable of adding subnetworks of different depths and widths to create a diverse ensemble, and trade off performance improvement with the number of parameters.
AdaNet adaptively growing an ensemble of neural networks. At each iteration, it measures the ensemble loss for each candidate, and selects the best one to move onto the next iteration.
Fast and Easy to Use
AdaNet implements the TensorFlow Estimator interface, which greatly simplifies machine learning programming by encapsulating training, evaluation, prediction and export for serving. It integrates with open-source tools like TensorFlow Hub modules, TensorFlow Model Analysis, and Google Cloud’s Hyperparameter Tuner. Distributed training support significantly reduces training time, and scales linearly with available CPUs and accelerators (e.g. GPUs).
AdaNet’s accuracy (y-axis) per train step (x-axis) on CIFAR-100. The blue line is accuracy on the training set, and red line is performance on the test set. A new subnetwork begins training every million steps, and eventually improves the performance of the ensemble. The grey and green lines are the accuracies of the ensemble before adding the new subnetwork.
Because TensorBoard is one of the best TensorFlow features for visualizing model metrics during training, AdaNet integrates seamlessly with it in order to monitor subnetwork training, ensemble composition, and performance. When AdaNet is done training, it exports a SavedModel that can be deployed with TensorFlow Serving.

Learning Guarantees
Building an ensemble of neural networks has several challenges: What are the best subnetwork architectures to consider? Is it best to reuse the same architectures or encourage diversity? While complex subnetworks with more parameters will tend to perform better on the training set, they may not generalize to unseen data due to their greater complexity. These challenges stem from evaluating model performance. We could evaluate performance on a hold-out set split from the training set, but in doing so would reduce the number of examples one can use for training the neural network.

Instead, AdaNet’s approach (presented in “AdaNet: Adaptive Structural Learning of Artificial Neural Networks” at ICML 2017) is to optimize an objective that balances the trade-offs between the ensemble’s performance on the training set and its ability to generalize to unseen data. The intuition is for the ensemble to include a candidate subnetwork only when it improves the ensemble’s training loss more than it affects its ability to generalize. This guarantees that:
  1. The generalization error of the ensemble is bounded by its training error and complexity.
  2. By optimizing this objective, we are directly minimizing this bound.
A practical benefit of optimizing this objective is that it eliminates the need for a hold-out set for choosing which candidate subnetworks to add to the ensemble. This has the added benefit of enabling the use of more training data for training the subnetworks. To learn more, please walk through our tutorial about the AdaNet objective.

Extensible
We believe that the key to making a useful AutoML framework for both research and production use is to not only provide sensible defaults, but to also allow users to try their own subnetwork/model definitions. As a result, machine learning researchers, practitioners, and enthusiasts are invited to define their own AdaNet adanet.subnetwork.Builder using high level TensorFlow APIs like tf.layers.

Users who have already integrated a TensorFlow model in their system can easily convert their TensorFlow code into an AdaNet subnetwork, and use the adanet.Estimator to boost model performance while obtaining learning guarantees. AdaNet will explore their defined search space of candidate subnetworks and learn to ensemble the subnetworks. For instance, we took an open-source implementation of a NASNet-A CIFAR architecture, transformed it into a subnetwork, and improved upon CIFAR-10 state-of-the-art results after eight AdaNet iterations. Furthermore, our model achieves this result with fewer parameters:
Performance of a NASNet-A model as presented in Zoph et al., 2018 versus AdaNet learning to combine small NASNet-A subnetworks on CIFAR-10.
Users are also invited to use their own custom loss functions as part of the AdaNet objective via canned or custom tf.contrib.estimator.Heads in order to train regression, classification, and multi-task learning problems.

Users can also fully define the search space of candidate subnetworks to explore by extending the adanet.subnetwork.Generator class. This allows them to grow or reduce their search space based on their available hardware. The search space of subnetworks can be as simple as duplicating the same subnetwork configuration with different random seeds, to training dozens of subnetworks with different hyperparameter combinations, and letting AdaNet choose the one to include in the final ensemble.

If you’re interested in trying AdaNet for yourself, please check out our Github repo, and walk through the tutorial notebooks. We’ve included a few working examples using dense layers and convolutions to get you started. AdaNet is an ongoing research project, and we welcome contributions. We’re excited to see how AdaNet can help the research community.

Acknowledgements
This project was only possible thanks to the members of the core team including Corinna Cortes, Mehryar Mohri, Xavi Gonzalvo, Charles Weill, Vitaly Kuznetsov, Scott Yak, and Hanna Mazzawi. We also extend a special thanks to our collaborators, residents and interns Gus Kristiansen, Galen Chuang, Ghassen Jerfel, Vladimir Macko, Ben Adlam, Scott Yang and the many others at Google who helped us test it out.

Source: Google AI Blog


Curiosity and Procrastination in Reinforcement Learning



Reinforcement learning (RL) is one of the most actively pursued research techniques of machine learning, in which an artificial agent receives a positive reward when it does something right, and negative reward otherwise. This carrot-and-stick approach is simple and universal, and allowed DeepMind to teach the DQN algorithm to play vintage Atari games and AlphaGoZero to play the ancient game of Go. This is also how OpenAI taught its OpenAI-Five algorithm to play the modern video game Dota, and how Google taught robotic arms to grasp new objects. However, despite the successes of RL, there are many challenges to making it an effective technique.

Standard RL algorithms struggle with environments where feedback to the agent is sparse — crucially, such environments are common in the real world. As an example, imagine trying to learn how to find your favorite cheese in a large maze-like supermarket. You search and search but the cheese section is nowhere to be found. If at every step you receive no “carrot” and no “stick”, there’s no way to tell if you are headed in the right direction or not. In the absence of rewards, what is to stop you from wandering around in circles? Nothing, except perhaps your curiosity, which motivates you go into a product section that looks unfamiliar to you in pursuit of your sought-after cheese.

In “Episodic Curiosity through Reachability” — the result of a collaboration between the Google Brain team, DeepMind and ETH Zürich — we propose a novel episodic memory-based model of granting RL rewards, akin to curiosity, which leads to exploring the environment. Since we want the agent not only to explore the environment but also to solve the original task, we add a reward bonus provided by our model to the original sparse task reward. The combined reward is not sparse anymore which allows standard RL algorithms to learn from it. Thus, our curiosity method expands the set of tasks which are solvable with RL.
Episodic Curiosity through Reachability: Observations are added to memory, reward is computed based on how far the current observation is from the most similar observation in memory. The agent receives more reward for seeing observations which are not yet represented in memory.
The key idea of our method is to store the agent's observations of the environment in an episodic memory, while also rewarding the agent for reaching observations not yet represented in memory. Being “not in memory” is the definition of novelty in our method — seeking such observations means seeking the unfamiliar. Such a drive to seek the unfamiliar will lead the artificial agent to new locations, thus keeping it from wandering in circles and ultimately help it stumble on the goal. As we will discuss later, our formulation can save the agent from undesired behaviours which some other formulations are prone to. Much to our surprise, those behaviours bear some similarity to what a layperson would call “procrastination”.

Previous Curiosity Formulations
While there have been many attempts to formulate curiosity in the past[1][2][3][4], in this post we  focus on one natural and very popular approach: curiosity through prediction-based surprise, explored in the recent paper “Curiosity-driven Exploration by Self-supervised Prediction” (commonly referred to as the ICM method). To illustrate how surprise leads to curiosity, again consider our analogy of looking for cheese in a supermarket.
Illustration © Indira Pasko, used under CC BY-NC-ND 4.0 license.
As you wander throughout the market, you try to predict the future (“Now I’m in the meat section, so I think the section around the corner is the fish section — those are usually adjacent in this supermarket chain”). If your prediction is wrong, you are surprised (“No, it’s actually the vegetables section. I didn’t expect that!”) and thus rewarded. This makes you more motivated to look around the corner in the future, exploring new locations just to see if your expectations about them meet the reality (and, hopefully, stumble upon the cheese).

Similarly, the ICM method builds a predictive model of the dynamics of the world and gives the agent rewards when the model fails to make good predictions — a marker of surprise or novelty. Note that exploring unvisited locations is not directly a part of the ICM curiosity formulation. For the ICM method, visiting them is only a way to obtain more “surprise” and thus maximize overall rewards. As it turns out, in some environments there could be other ways to inflict self-surprise, leading to unforeseen results.
Agent imbued with surprise-based curiosity gets stuck when it encounters TV. GIF adopted from a video by © Deepak Pathak, used under CC BY 2.0 license.
The Dangers of “Procrastination”
In "Large-Scale Study of Curiosity-Driven Learning", the authors of the ICM method along with researchers from OpenAI show a hidden danger of surprise maximization: agents can learn to indulge procrastination-like behaviour instead of doing something useful for the task at hand. To see why, consider a common thought experiment the authors call the “noisy TV problem”, in which an agent is put into a maze and tasked with finding a highly rewarding item (akin to “cheese” in our previous supermarket example). The environment also contains a TV for which the agent has the remote control. There is a limited number of channels (each with a distinct show) and every press on the remote control switches to a random channel. How would an agent perform in such an environment?

For the surprise-based curiosity formulation, changing channels would result in a large reward, as each change is unpredictable and surprising. Crucially, even after cycling through all the available channels, the random channel selection ensures every new change will still be surprising — the agent is making predictions about what will be on the TV after a channel change, and will very likely be wrong, leading to surprise. Importantly, even if the agent has already seen every show on every channel, the change is still unpredictable. Because of this, the agent imbued with surprise-based curiosity would eventually stay in front of the TV forever instead of searching for a highly rewarding item — akin to procrastination. So, what would be a definition of curiosity which does not lead to such behaviour?

Episodic Curiosity
In “Episodic Curiosity through Reachability”, we explore an episodic memory-based curiosity model that turns out to be less prone to “self-indulging” instant gratification. Why so? Using our example above, after changing channels for a while, all of the shows will end up in memory. Thus, the TV won’t be so attractive anymore: even if the order of shows appearing on the screen is random and unpredictable, all those shows are already in memory! This is the main difference to the surprise-based methods: our method doesn’t even try to make bets about the future which could be hard (or even impossible) to predict. Instead, the agent examines the past to know if it has seen observations similar to the current one. Thus our agent won’t be drawn that much to the instant gratification provided by the noisy TV. It will have to go and explore the world outside of the TV to get more reward.

But how do we decide whether the agent is seeing the same thing as an existing memory? Checking for an exact match could be meaningless: in a realistic environment, the agent rarely sees exactly the same thing twice. For example, even if the agent returned to exactly the same room, it would still see this room under a different angle compared to its memories.

Instead of checking for an exact match in memory, we use a deep neural network that is trained to measure how similar two experiences are. To train this network, we have it guess whether two observations were experienced close together in time, or far apart in time. Temporal proximity is a good proxy for whether two experiences should be judged to be part of the same experience. This training leads to a general concept of novelty via reachability which is illustrated below.
Graph of reachabilities would determine novelty. In practice, this graph is not available — so we train a neural network approximator to estimate a number of steps between observations.
Experimental Results
To compare the performance of different approaches to curiosity, we tested them in two visually rich 3D environments: ViZDoom and DMLab. In those environments, the agent was tasked with various problems like searching for a goal in a maze or collecting good and avoiding bad objects. The DMLab environment happens to provide the agent with a laser-like science fiction gadget. The standard setting in the previous work on DMLab was to equip the agent with this gadget for all tasks, and if the agent does not need a gadget for a particular task, it is free not to use it. Interestingly, similar to the noisy TV experiment described above, the surprise-based ICM method actually uses this gadget a lot even when it is useless for the task at hand! When tasked with searching for a high-rewarding item in the maze, it instead prefers to spend time tagging walls because this yields a lot of “surprise” reward. Theoretically, predicting the result of tagging should be possible, but in practice is too hard as it apparently requires a deeper knowledge of physics than is available to a standard agent.
Surprise-based ICM method is persistently tagging the wall instead of exploring the maze.
Our method instead learns reasonable exploration behaviour under the same conditions. This is because it does not try to predict the result of its actions, but rather seeks observations which are “harder” to achieve from those already in the episodic memory. In other words, the agent implicitly pursues goals which require more effort to reach from memory than just a single tagging action.
Our method shows reasonable exploration.
It is interesting to see that our approach to granting reward penalizes an agent running in circles. This is because after completing the first circle the agent does not encounter new observations other than those in memory, and thus receives no reward:
Our reward visualization: red means negative reward, green means positive reward. Left to right: map with rewards, map with locations currently in memory, first-person view.
At the same time, our method favors good exploration behavior:
Our reward visualization: red means negative reward, green means positive reward. Left to right: map with rewards, map with locations currently in memory, first-person view.
We hope that our work will help lead to a new wave of exploration methods, going beyond surprise and learning more intelligent exploration behaviours. For an in-depth analysis of our method, please take a look at the preprint of our research paper.

Acknowledgements:
This project is a result of a collaboration between the Google Brain team, DeepMind and ETH Zürich. The core team includes Nikolay Savinov, Anton Raichuk, Raphaël Marinier, Damien Vincent, Marc Pollefeys, Timothy Lillicrap and Sylvain Gelly. We would like to thank Olivier Pietquin, Carlos Riquelme, Charles Blundell and Sergey Levine for the discussions about the paper. We are grateful to Indira Pasko for the help with illustrations.

References:
[1] "Count-Based Exploration with Neural Density Models", Georg Ostrovski, Marc G. Bellemare, Aaron van den Oord, Remi Munos
[2] "#Exploration: A Study of Count-Based Exploration for Deep Reinforcement Learning", Haoran Tang, Rein Houthooft, Davis Foote, Adam Stooke, Xi Chen, Yan Duan, John Schulman, Filip De Turck, Pieter Abbeel
[3] "Unsupervised Learning of Goal Spaces for Intrinsically Motivated Goal Exploration", Alexandre Péré, Sébastien Forestier, Olivier Sigaud, Pierre-Yves Oudeyer
[4] "VIME: Variational Information Maximizing Exploration", Rein Houthooft, Xi Chen, Yan Duan, John Schulman, Filip De Turck, Pieter Abbeel

Source: Google AI Blog


Fluid Annotation: An Exploratory Machine Learning–Powered Interface for Faster Image Annotation



The performance of modern deep learning–based computer vision models, such as those implemented by the TensorFlow Object Detection API, depends on the availability of increasingly large, labeled training datasets, such as Open Images. However, obtaining high-quality training data is quickly becoming a major bottleneck in computer vision. This is especially the case for pixel-wise prediction tasks such as semantic segmentation, used in applications such as autonomous driving, robotics, and image search. Indeed, traditional manual labeling tools require an annotator to carefully click on the boundaries to outline each object in the image, which is tedious: labeling a single image in the COCO+Stuff dataset takes 19 minutes, while labeling the whole dataset would take over 53k hours!
Example of image in the COCO dataset (left) and its pixel-wise semantic labeling (right). Image credit: Florida Memory, original image.
In “Fluid Annotation: A Human-Machine Collaboration Interface for Full Image Annotation”, to be presented at the Brave New Ideas track of the 2018 ACM Multimedia Conference, we explore a machine learning–powered interface for annotating the class label and outline of every object and background region in an image, accelerating the creation of labeled datasets by a factor of 3x.

Fluid Annotation starts from the output of a strong semantic segmentation model, which a human annotator can modify through machine-assisted edit operations using a natural user interface. Our interface empowers annotators to choose what to correct and in which order, allowing them to effectively focus their efforts on what the machine does not already know.
Visualization of the fluid annotation interface in action on image from COCO dataset. Image credit: gamene, original image.
More precisely, to annotate an image we first run it through a pre-trained semantic segmentation model (Mask-RCNN). This generates around 1000 image segments with their class labels and confidence scores. The segments with the highest confidences are used to initialize the labeling which is presented to the annotator. Afterwards, the annotator can: (1) Change the label of an existing segment choosing from a shortlist generated by the machine. (2) Add a segment to cover a missing object. The machine identifies the most likely pre-generated segments, through which the annotator can scroll and select the best one. (3) Remove an existing segment. (4) Change the depth-order of overlapping segments. To get a better feeling for this interface, try out the demo (desktop only).
Comparison of annotations using traditional manual labeling tools (middle column) and fluid annotation (right) on three COCO images. While object boundaries are often more accurate when using manual labeling tools, the biggest source of annotation differences is because human annotators often disagree on the exact object class. Image Credits: sneaka, original image (top), Dan Hurt, original image (middle), Melodie Mesiano, original image (bottom).
Fluid Annotation is a first exploratory step towards making image annotation faster and easier. In future work we aim to improve the annotation of object boundaries, make the interface faster by including more machine intelligence, and finally extend the interface to handle previous unseen classes for which efficient data collection is needed the most.

Acknowledgements
This work was done in collaboration with Misha Andriluka. Special thanks to Christine Sugrue for creating the fluid annotation demo. We also thank Anna Ukhanova and Damien Henry for their valuable input.

Source: Google AI Blog