dennybritz / deeplearning-papernotes

Summaries and notes on Deep Learning research papers

2018-02

• The Matrix Calculus You Need For Deep Learning [arXiv]
• Regularized Evolution for Image Classifier Architecture Search [arXiv]
• Online Learning: A Comprehensive Survey [arXiv]
• Visual Interpretability for Deep Learning: a Survey [arXiv]
• Behavior is Everything – Towards Representing Concepts with Sensorimotor Contingencies [paper] [article] [code]
• IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures [arXiv] [article] [code]
• DeepType: Multilingual Entity Linking by Neural Type System Evolution [arXiv] [article] [code]
• DensePose: Dense Human Pose Estimation In The Wild [arXiv] [article]

2018-01

• Nested LSTMs [arXiv]
• Generating Wikipedia by Summarizing Long Sequences [arXiv]
• Scalable and accurate deep learning for electronic health records [arXiv]
• Kernel Feature Selection via Conditional Covariance Minimization [NIPS paper] [article] [code]
• Psychlab: A Psychology Laboratory for Deep Reinforcement Learning Agents [arXiv] [article] [code]
• Fine-tuned Language Models for Text Classification [arXiv] [code] (soon)
• Deep Learning: An Introduction for Applied Mathematicians [arXiv]
• Innateness, AlphaZero, and Artificial Intelligence [arXiv]
• Can Computers Create Art? [arXiv]
• eCommerceGAN : A Generative Adversarial Network for E-commerce [arXiv]
• Expected Policy Gradients for Reinforcement Learning [arXiv]
• DroNet: Learning to Fly by Driving [UZH docs] [article] [code]
• Symmetric Decomposition of Asymmetric Games [Scientific Reports] [article]
• Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor [arXiv] [code]
• SBNet: Sparse Blocks Network for Fast Inference [arXiv] [article] [code]
• DeepMind Control Suite [arXiv] [code]
• Deep Learning: A Critical Appraisal [arXiv]

2017-12

• Adversarial Patch [arXiv]
• CNN Is All You Need [arXiv]
• Learning Robot Objectives from Physical Human Interaction [paper] [article]
• The NarrativeQA Reading Comprehension Challenge [arXiv] [dataset]
• Objects that Sound [arXiv]
• Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions [arXiv] [article] [article2]
• Deep Neuroevolution: Genetic Algorithms Are a Competitive Alternative for Training Deep Neural Networks for Reinforcement Learning [arXiv] [article] [code]
• Improving Exploration in Evolution Strategies for Deep Reinforcement Learning via a Population of Novelty-Seeking Agents [arXiv] [article] [code]
• Superhuman AI for heads-up no-limit poker: Libratus beats top professionals [Science]
• Mathematics of Deep Learning [arXiv]
• State-of-the-art Speech Recognition With Sequence-to-Sequence Models [arXiv] [article]
• Peephole: Predicting Network Performance Before Training [arXiv]
• Deliberation Network: Pushing the frontiers of neural machine translation [Research at Microsoft] [article]
• GPU Kernels for Block-Sparse Weights [Research at OpenAI] [article] [code]
• Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm [arXiv]
• Deep Learning Scaling is Predictable, Empirically [arXiv] [article]

2017-11

• High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs [arXiv] [article] [code]
• StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation [arXiv] [code]
• Population Based Training of Neural Networks [arXiv] [article]
• Distilling a Neural Network Into a Soft Decision Tree [arXiv]
• Neural Text Generation: A Practical Guide [arXiv]
• Parallel WaveNet: Fast High-Fidelity Speech Synthesis [DeepMind documents] [article]
• CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning [arXiv] [article]
• Non-local Neural Networks [arXiv]
• Deep Image Prior [paper] [article] [code]
• Online Deep Learning: Learning Deep Neural Networks on the Fly [arXiv]
• Learning Explanatory Rules from Noisy Data [arXiv]
• Improving Palliative Care with Deep Learning [arXiv] [article]
• VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection [arXiv]
• Weighted Transformer Network for Machine Translation [arXiv] [article]
• Non-Autoregressive Neural Machine Translation [arXiv] [article]
• Block-Sparse Recurrent Neural Networks [arXiv]
• A Unified Game-Theoretic Approach to Multiagent Reinforcement Learning [arXiv]
• Neural Discrete Representation Learning [arXiv] [article]
• Don't Decay the Learning Rate, Increase the Batch Size [arXiv]
• Hierarchical Representations for Efficient Architecture Search [arXiv]

2017-10

• Unsupervised Machine Translation Using Monolingual Corpora Only [arXiv]
• Dynamic Routing Between Capsules [arXiv]
• A generative vision model that trains with high data efficiency and breaks text-based CAPTCHAs [Science] [article] [code]
• Understanding Grounded Language Learning Agents [arXiv]
• Planning, Fast and Slow: A Framework for Adaptive Real-Time Safe Trajectory Planning [arXiv] [article] [code] (soon)
• Malware Detection by Eating a Whole EXE [arXiv] [article]
• Progressive Growing of GANs for Improved Quality, Stability, and Variation [Research at Nvidia] [article] [code]
• Meta Learning Shared Hierarchies [arXiv] [article] [code]
• Deep Voice 3: 2000-Speaker Neural Text-to-Speech [arXiv] [article]
• AVA: A Video Dataset of Spatio-temporally Localized Atomic Visual Actions [arXiv] [article] [dataset]
• Mastering the game of Go without Human Knowledge [Nature] [article]
• Sim-to-Real Transfer of Robotic Control with Dynamics Randomization [arXiv] [article]
• Asymmetric Actor Critic for Image-Based Robot Learning [arXiv] [article]
• A systematic study of the class imbalance problem in convolutional neural networks [arXiv]
• Generalization in Deep Learning [arXiv]
• Swish: a Self-Gated Activation Function [arXiv]
• Emergent Translation in Multi-Agent Communication [arXiv]
• SLING: A framework for frame semantic parsing [arXiv] [article] [code]
• Meta-Learning for Wrestling [arXiv] [article] [code]
• Mixed Precision Training [arXiv] [article] [article2] [code/docs]
• Generative Adversarial Networks: An Overview [arXiv]
• Emergent Complexity via Multi-Agent Competition [arXiv] [article] [code]
• Deep Lattice Networks and Partial Monotonic Functions [Research at Google] [article] [code]
• The IIT Bombay English-Hindi Parallel Corpus [arXiv] [article]
• Rainbow: Combining Improvements in Deep Reinforcement Learning [arXiv]
• Lifelong Learning With Dynamically Expandable Networks [arXiv]
• Variational Inference & Deep Learning: A New Synthesis (Thesis) [dropbox]
• Neural Task Programming: Learning to Generalize Across Hierarchical Tasks [arXiv]
• Neural Color Transfer between Images [arXiv]
• The hippocampus as a predictive map [biorXiv] [article]
• Scalable and accurate deep learning for electronic health records [arXiv]

2017-09

• Variational Memory Addressing in Generative Models [arXiv]
• Overcoming Exploration in Reinforcement Learning with Demonstrations [arXiv]
• A Hybrid DSP/Deep Learning Approach to Real-Time Full-Band Speech Enhancement [arXiv] [article] [code]
• ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases [CVF] [article] [dataset]
• NIMA: Neural Image Assessment [arXiv] [article]
• Generating Sentences by Editing Prototypes [arXiv] [code]
• The Consciousness Prior [arXiv]
• StarSpace: Embed All The Things! [arXiv] [code]
• Neural Optimizer Search with Reinforcement Learning [arXiv]
• Dynamic Evaluation of Neural Sequence Models [arXiv]
• Neural Machine Translation [arXiv]
• Matterport3D: Learning from RGB-D Data in Indoor Environments [arXiv] [article] [article2] [code]
• Deep Reinforcement Learning that Matters [arXiv] [code]
• The Uncertainty Bellman Equation and Exploration [arXiv]
• WESPE: Weakly Supervised Photo Enhancer for Digital Cameras [arXiv] [article]
• Globally Normalized Reader [arXiv] [article] [code]
• A Brief Introduction to Machine Learning for Engineers [arXiv]
• Learning with Opponent-Learning Awareness [arXiv] [article]
• A Deep Reinforcement Learning Chatbot [arXiv]
• Squeeze-and-Excitation Networks [arXiv]
• Efficient Methods and Hardware for Deep Learning (Thesis) [Stanford Digital Repository]

2017-08

• Design and Analysis of the NIPS 2016 Review Process [arXiv]
• Fast Automated Analysis of Strong Gravitational Lenses with Convolutional Neural Networks [arXiv] [article]
• TensorFlow Agents: Efficient Batched Reinforcement Learning in TensorFlow [white paper] [code]
• Automated Crowdturfing Attacks and Defenses in Online Review Systems [arXiv]
• Neural Network Dynamics for Model-Based Deep Reinforcement Learning with Model-Free Fine-Tuning [arXiv] [article] [code]
• Deep Learning for Video Game Playing [arXiv]
• Deep & Cross Network for Ad Click Predictions [arXiv]
• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms [arXiv] [code]
• Multi-task Self-Supervised Visual Learning [arXiv]
• Learning a Multi-View Stereo Machine [arXiv] [article] [code] (soon)
• Twin Networks: Using the Future as a Regularizer [arXiv]
• A Brief Survey of Deep Reinforcement Learning [arXiv]
• Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation [arXiv] [code]
• On the Effectiveness of Visible Watermarks [CVPR] [article]
• Practical Network Blocks Design with Q-Learning [arXiv]
• On Ensuring that Intelligent Machines Are Well-Behaved [arXiv]
• Reproducibility of Benchmarked Deep Reinforcement Learning Tasks for Continuous Control [arXiv] [code]
• Training Deep AutoEncoders for Collaborative Filtering [arXiv] [code]
• Learning to Perform a Perched Landing on the GroundUsing Deep Reinforcement Learning [nature]
• Revisiting the Effectiveness of Off-the-shelf Temporal Modeling Approaches for Large-scale Video Classification [arXiv] [article]
• Intrinsically Motivated Goal Exploration Processes with Automatic Curriculum Learning [arXiv]
• Neural Expectation Maximization [arXiv] [code]
• Google Vizier: A Service for Black-Box Optimization [Research at Google]
• STARDATA: A StarCraft AI Research Dataset [arXiv] [code]
• Using millions of emoji occurrences to learn any-domain representations for detecting sentiment, emotion and sarcasm [arXiv] [code] [article]
• Natural Language Processing with Small Feed-Forward Networks [arXiv]

2017-07

• Photographic Image Synthesis with Cascaded Refinement Networks [arXiv] [code]
• StarCraft II: A New Challenge for Reinforcement Learning [DeepMind Documents] [code] [article]
• Leveraging Demonstrations for Deep Reinforcement Learning on Robotics Problems with Sparse Rewards [arXiv]
• Reinforcement Learning with Deep Energy-Based Policies [arXiv] [article] [code]
• DARLA: Improving Zero-Shot Transfer in Reinforcement Learning [arXiv]
• Synthesizing Robust Adversarial Examples [arXiv] [article] [code] (Soon)
• Voice Synthesis for in-the-Wild Speakers via a Phonological Loop [arXiv] [code] [article]
• Eyemotion: Classifying facial expressions in VR using eye-tracking cameras [arXiv] [article]
• A Distributional Perspective on Reinforcement Learning [arXiv] [article] [video]
• On the State of the Art of Evaluation in Neural Language Models [arXiv]
• Optimizing the Latent Space of Generative Networks [arXiv]
• Neuroscience-Inspired Artificial Intelligence [Neuron] [article]
• Learning Transferable Architectures for Scalable Image Recognition [arXiv]
• Reverse Curriculum Generation for Reinforcement Learning [arXiv]
• Imagination-Augmented Agents for Deep Reinforcement Learning [arXiv] [article]
• Learning model-based planning from scratch [arXiv] [article]
• Proximal Policy Optimization Algorithms [AWSS3] [code]
• Automatic Recognition of Deceptive Facial Expressions of Emotion [arXiv]
• Distral: Robust Multitask Reinforcement Learning [arXiv]
• Creatism: A deep-learning photographer capable of creating professional work [arXiv] [article]
• SCAN: Learning Abstract Hierarchical Compositional Visual Concepts [arXiv] [article]
• Revisiting Unreasonable Effectiveness of Data in Deep Learning Era [arXiv] [article]
• The Intentional Unintentional Agent: Learning to Solve Many Continuous Control Tasks Simultaneously [arXiv]
• Deep Bilateral Learning for Real-Time Image Enhancement [arXiv] [code] [article]
• Emergence of Locomotion Behaviours in Rich Environments [arXiv] [article]
• Learning human behaviors from motion capture by adversarial imitation [arXiv] [article]
• Robust Imitation of Diverse Behaviors [arXiv] [article]
• Hindsight Experience Replay [arXiv]
• Cardiologist-Level Arrhythmia Detection with Convolutional Neural Networks [arXiv] [article]
• End-to-End Learning of Semantic Grasping [arXiv]
• ELF: An Extensive, Lightweight and Flexible Research Platform for Real-time Strategy Games [arXiv] [code] [article]

2017-06

• Noisy Networks for Exploration [arXiv]
• Do GANs actually learn the distribution? An empirical study [arXiv]
• Gradient Episodic Memory for Continuum Learning [arXiv]
• Natural Language Does Not Emerge 'Naturally' in Multi-Agent Dialog [arXiv] [code]
• Deep Interest Network for Click-Through Rate Prediction [arXiv]
• Cognitive Psychology for Deep Neural Networks: A Shape Bias Case Study [arXiv] [article]
• Structure Learning in Motor Control: A Deep Reinforcement Learning Model [arXiv]
• Programmable Agents [arXiv]
• Grounded Language Learning in a Simulated 3D World [arXiv]
• Schema Networks: Zero-shot Transfer with a Generative Causal Model of Intuitive Physics [arXiv]
• SVCCA: Singular Vector Canonical Correlation Analysis for Deep Learning Dynamics and Interpretability [arXiv] [article] [code]
• One Model To Learn Them All [arXiv] [code] [article]
• Hybrid Reward Architecture for Reinforcement Learning [arXiv]
• Expected Policy Gradients [arXiv]
• Variational Approaches for Auto-Encoding Generative Adversarial Networks [arXiv]
• Deal or No Deal? End-to-End Learning for Negotiation Dialogues [S3AWS] [code] [article]
• Attention Is All You Need [arXiv] [code] [article]
• Sobolev Training for Neural Networks [arXiv]
• YellowFin and the Art of Momentum Tuning [arXiv] [code] [article]
• Forward Thinking: Building and Training Neural Networks One Layer at a Time [arXiv]
• Depthwise Separable Convolutions for Neural Machine Translation [arXiv] [code]
• Parameter Space Noise for Exploration [arXiv] [code] [article]
• Deep Reinforcement Learning from human preferences [arXiv] [article]
• Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments [arXiv] [code]
• Self-Normalizing Neural Networks [arXiv] [code]
• Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour [arXiv]
• A simple neural network module for relational reasoning [arXiv] [article]
• Visual Interaction Networks [arXiv] [article]

2017-05

• Supervised Learning of Universal Sentence Representations from Natural Language Inference Data [arXiv] [code]
• pix2code: Generating Code from a Graphical User Interface Screenshot [arXiv] [article] [code]
• The Cramer Distance as a Solution to Biased Wasserstein Gradients [arXiv]
• Reinforcement Learning with a Corrupted Reward Channel [arXiv]
• Dilated Residual Networks [arXiv] [code]
• Bayesian GAN [arXiv] [code]
• Gradient Descent Can Take Exponential Time to Escape Saddle Points [arXiv] [article]
• Thinking Fast and Slow with Deep Learning and Tree Search [arXiv]
• ParlAI: A Dialog Research Software Platform [arXiv] [code] [article]
• Semantically Decomposing the Latent Spaces of Generative Adversarial Networks [arXiv] [article]
• Look, Listen and Learn [arXiv]
• Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset [arXiv] [code]
• Convolutional Sequence to Sequence Learning [arXiv] [code] [code2] [article]
• The Kinetics Human Action Video Dataset [arXiv] [article]
• Safe and Nested Subgame Solving for Imperfect-Information Games [arXiv]
• Discrete Sequential Prediction of Continuous Actions for Deep RL [arXiv]
• Metacontrol for Adaptive Imagination-Based Optimization [arXiv]
• Efficient Parallel Methods for Deep Reinforcement Learning [arXiv]
• Real-Time Adaptive Image Compression [arXiv]

2017-04

• General Video Game AI: Learning from Screen Capture [arXiv]
• Learning to Skim Text [arXiv]
• Get To The Point: Summarization with Pointer-Generator Networks [arXiv] [code] [article]
• Adversarial Neural Machine Translation [arXiv]
• Deep Q-learning from Demonstrations [arXiv]
• Learning from Demonstrations for Real World Reinforcement Learning [arXiv]
• DSLR-Quality Photos on Mobile Devices with Deep Convolutional Networks [arXiv] [article] [code]
• A Neural Representation of Sketch Drawings [arXiv] [code] [article]
• Automated Curriculum Learning for Neural Networks [arXiv]
• Hierarchical Surface Prediction for 3D Object Reconstruction [arXiv] [article]
• Neural Message Passing for Quantum Chemistry [arXiv]
• Learning to Generate Reviews and Discovering Sentiment [arXiv] [code]
• Best Practices for Applying Deep Learning to Novel Applications [arXiv]

2017-03

• Improved Training of Wasserstein GANs [arXiv]
• Evolution Strategies as a Scalable Alternative to Reinforcement Learning [arXiv]
• Controllable Text Generation [arXiv]
• Neural Episodic Control [arXiv]
• A Structured Self-attentive Sentence Embedding [arXiv]
• Multi-step Reinforcement Learning: A Unifying Algorithm [arXiv]
• Deep learning with convolutional neural networks for brain mapping and decoding of movement-related information from the human EEG [arXiv]
• FaSTrack: a Modular Framework for Fast and Guaranteed Safe Motion Planning [arXiv] [article] [article2]
• Massive Exploration of Neural Machine Translation Architectures [arXiv] [code]
• Large Pose 3D Face Reconstruction from a Single Image via Direct Volumetric CNN Regression [arXiv] [article] [code]
• Minimax Regret Bounds for Reinforcement Learning [arXiv]
• Sharp Minima Can Generalize For Deep Nets [arXiv]
• Parallel Multiscale Autoregressive Density Estimation [arXiv]
• Neural Machine Translation and Sequence-to-sequence Models: A Tutorial [arXiv]
• Large-Scale Evolution of Image Classifiers [arXiv]
• FeUdal Networks for Hierarchical Reinforcement Learning [arXiv]
• Evolving Deep Neural Networks [arXiv]
• How to Escape Saddle Points Efficiently [arXiv] [article]
• Opening the Black Box of Deep Neural Networks via Information [arXiv] [video]
• Understanding Synthetic Gradients and Decoupled Neural Interfaces [arXiv]
• Learning to Optimize Neural Nets [arXiv] [article]

2017-02

• The Shattered Gradients Problem: If resnets are the answer, then what is the question? [arXiv]
• Neural Map: Structured Memory for Deep Reinforcement Learning [arXiv]
• Bridging the Gap Between Value and Policy Based Reinforcement Learning [arXiv]
• Deep Voice: Real-time Neural Text-to-Speech [arXiv]
• Beating the World's Best at Super Smash Bros. with Deep Reinforcement Learning [arXiv]
• The Game Imitation: Deep Supervised Convolutional Networks for Quick Video Game AI [arXiv]
• Learning to Parse and Translate Improves Neural Machine Translation [arXiv]
• All-but-the-Top: Simple and Effective Postprocessing for Word Representations [arXiv]
• Deep Learning with Dynamic Computation Graphs [arXiv]
• Skip Connections as Effective Symmetry-Breaking [arXiv]
• odelSemi-Supervised QA with Generative Domain-Adaptive Nets [arXiv]

2017-01

• Wasserstein GAN [arXiv]
• Deep Reinforcement Learning: An Overview [arXiv]
• DyNet: The Dynamic Neural Network Toolkit [arXiv]
• DeepStack: Expert-Level Artificial Intelligence in No-Limit Poker [arXiv]
• NIPS 2016 Tutorial: Generative Adversarial Networks [arXiv]

2016-12

• A recurrent neural network without Chaos [arXiv]
• Language Modeling with Gated Convolutional Networks [arXiv]
• EnhanceNet: Single Image Super-Resolution Through Automated Texture Synthesis [arXiv] [article]
• Learning from Simulated and Unsupervised Images through Adversarial Training [arXiv]
• How Grammatical is Character-level Neural Machine Translation? Assessing MT Quality with Contrastive Translation Pairs [arXiv]
• Improving Neural Language Models with a Continuous Cache [arXiv]
• DeepMind Lab [arXiv] [code]
• Deep Learning of Robotic Tasks without a Simulator using Strong and Weak Human Supervision [arXiv]
• Knowing When to Look: Adaptive Attention via A Visual Sentinel for Image Captioning [arXiv]
• Overcoming catastrophic forgetting in neural networks [arXiv]

2016-11 (ICLR Edition)

• Image-to-Image Translation with Conditional Adversarial Networks [arXiv]
• Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer [OpenReview]
• Learning to reinforcement learn [arXiv]
• A Way out of the Odyssey: Analyzing and Combining Recent Insights for LSTMs [arXiv]
• Adversarial Training Methods for Semi-Supervised Text Classification [arXiv]
• Importance Sampling with Unequal Support [arXiv]
• Quasi-Recurrent Neural Networks [arXiv]
• Capacity and Learnability in Recurrent Neural Networks [OpenReview]
• Unrolled Generative Adversarial Networks [OpenReview]
• Deep Information Propagation [OpenReview]
• Structured Attention Networks [OpenReview]
• Incremental Sequence Learning [arXiv]
• Delving into Transferable Adversarial Examples and Black-box Attacks [arXiv] [code]
• b-GAN: Unified Framework of Generative Adversarial Networks [OpenReview]
• A Joint Many-Task Model: Growing a Neural Network for Multiple NLP Tasks [OpenReview]
• Categorical Reparameterization with Gumbel-Softmax [arXiv]
• Lip Reading Sentences in the Wild [arXiv]

Reinforcement Learning:

-Learning to reinforcement learn [arXiv]

• A Connection between Generative Adversarial Networks, Inverse Reinforcement Learning, and Energy-Based Models [arXiv]
• The Predictron: End-To-End Learning and Planning [OpenReview]
• Third-Person Imitation Learning [OpenReview]
• Generalizing Skills with Semi-Supervised Reinforcement Learning [OpenReview]
• Sample Efficient Actor-Critic with Experience Replay [OpenReview]
• Reinforcement Learning with Unsupervised Auxiliary Tasks [arXiv]
• Neural Architecture Search with Reinforcement Learning [OpenReview]
• Towards Information-Seeking Agents [OpenReview]
• Multi-Agent Cooperation and the Emergence of (Natural) Language [OpenReview]
• Improving Policy Gradient by Exploring Under-appreciated Rewards [OpenReview]
• Stochastic Neural Networks for Hierarchical Reinforcement Learning [OpenReview]
• Tuning Recurrent Neural Networks with Reinforcement Learning [OpenReview]
• RL^2: Fast Reinforcement Learning via Slow Reinforcement Learning [arXiv]
• Learning Invariant Feature Spaces to Transfer Skills with Reinforcement Learning [OpenReview]
• Learning to Perform Physics Experiments via Deep Reinforcement Learning [OpenReview]
• Reinforcement Learning through Asynchronous Advantage Actor-Critic on a GPU [OpenReview]
• Learning to Compose Words into Sentences with Reinforcement Learning[OpenReview]
• Deep Reinforcement Learning for Accelerating the Convergence Rate [OpenReview]
• #Exploration: A Study of Count-Based Exploration for Deep Reinforcement Learning [arXiv]
• Learning to Compose Words into Sentences with Reinforcement Learning [OpenReview]
• Learning to Navigate in Complex Environments [arXiv]
• Unsupervised Perceptual Rewards for Imitation Learning [OpenReview]
• Q-Prop: Sample-Efficient Policy Gradient with An Off-Policy Critic [OpenReview]

Machine Translation & Dialog

• Google's Multilingual Neural Machine Translation System: Enabling Zero-Shot Translation [arXiv]
• Neural Machine Translation with Reconstruction [arXiv]
• Iterative Refinement for Machine Translation [OpenReview]
• A Convolutional Encoder Model for Neural Machine Translation [arXiv]
• Improving Neural Language Models with a Continuous Cache [OpenReview]
• Vocabulary Selection Strategies for Neural Machine Translation [OpenReview]
• Towards an automatic Turing test: Learning to evaluate dialogue responses [OpenReview]
• Dialogue Learning With Human-in-the-Loop [OpenReview]
• Batch Policy Gradient Methods for Improving Neural Conversation Models [OpenReview]
• Learning through Dialogue Interactions [OpenReview]
• Dual Learning for Machine Translation [arXiv]
• Unsupervised Pretraining for Sequence to Sequence Learning [arXiv]

2016-10

• Hybrid computing using a neural network with dynamic external memory [nature] [code]
• Quantum Machine Learning [arXiv]
• Understanding deep learning requires rethinking generalization [arXiv]
• Universal adversarial perturbations [arXiv] [code]
• Neural Machine Translation in Linear Time [arXiv] [code]
• Professor Forcing: A New Algorithm for Training Recurrent Networks [arXiv]
• Learning to Protect Communications with Adversarial Neural Cryptography [arXiv]
• Can Active Memory Replace Attention? [arXiv]
• Using Fast Weights to Attend to the Recent Past [arXiv]
• Fully Character-Level Neural Machine Translation without Explicit Segmentation [arXiv]
• Diverse Beam Search: Decoding Diverse Solutions from Neural Sequence Models [arXiv]
• Video Pixel Networks [arXiv]
• Connecting Generative Adversarial Networks and Actor-Critic Methods [arXiv]
• Learning to Translate in Real-time with Neural Machine Translation [arXiv]
• Xception: Deep Learning with Depthwise Separable Convolutions [arXiv]
• Collective Robot Reinforcement Learning with Distributed Asynchronous Guided Policy Search [arXiv]
• Pointer Sentinel Mixture Models [arXiv]

2016-09

• Towards Deep Symbolic Reinforcement Learning [arXiv]
• HyperNetworks [arXiv]
• Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation [arXiv]
• Safe and Efficient Off-Policy Reinforcement Learning [arXiv]
• Playing FPS Games with Deep Reinforcement Learning [arXiv]
• SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient [arXiv]
• Episodic Exploration for Deep Deterministic Policies: An Application to StarCraft Micromanagement Tasks [arXiv]
• Energy-based Generative Adversarial Network [arXiv]
• Stealing Machine Learning Models via Prediction APIs [arXiv]
• Semi-Supervised Classification with Graph Convolutional Networks [arXiv]
• WaveNet: A Generative Model For Raw Audio [arXiv]
• Hierarchical Multiscale Recurrent Neural Networks [arXiv]
• End-to-End Reinforcement Learning of Dialogue Agents for Information Access [arXiv]
• Deep Neural Networks for YouTube Recommendations [paper]

2016-08

• Semantics derived automatically from language corpora contain human-like biases [arXiv]
• Why does deep and cheap learning work so well? [arXiv]
• Machine Comprehension Using Match-LSTM and Answer Pointer [arXiv]
• Stacked Approximated Regression Machine: A Simple Deep Learning Approach [arXiv]
• Decoupled Neural Interfaces using Synthetic Gradients [arXiv]
• WikiReading: A Novel Large-scale Language Understanding Task over Wikipedia [arXiv]
• Temporal Attention Model for Neural Machine Translation [arXiv]
• Residual Networks of Residual Networks: Multilevel Residual Networks [arXiv]
• Learning Online Alignments with Continuous Rewards Policy Gradient [arXiv]

2016-07

• An Actor-Critic Algorithm for Sequence Prediction [arXiv]
• Cognitive Science in the era of Artificial Intelligence: A roadmap for reverse-engineering the infant language-learner [arXiv]
• Recurrent Neural Machine Translation [arXiv]
• MS-Celeb-1M: A Dataset and Benchmark for Large-Scale Face Recognition [arXiv]
• Layer Normalization [arXiv]
• Neural Machine Translation with Recurrent Attention Modeling [arXiv]
• Neural Semantic Encoders [arXiv]
• Attention-over-Attention Neural Networks for Reading Comprehension [arXiv]
• sk_p: a neural program corrector for MOOCs [arXiv]
• Recurrent Highway Networks [arXiv]
• Bag of Tricks for Efficient Text Classification [arXiv]
• Context-Dependent Word Representation for Neural Machine Translation [arXiv]
• Dynamic Neural Turing Machine with Soft and Hard Addressing Schemes [arXiv]

2016-06

• Sequence-to-Sequence Learning as Beam-Search Optimization [arXiv]
• Sequence-Level Knowledge Distillation [arXiv]
• Policy Networks with Two-Stage Training for Dialogue Systems [arXiv]
• Towards an integration of deep learning and neuroscience [arXiv]
• On Multiplicative Integration with Recurrent Neural Networks [arxiv]
• Wide & Deep Learning for Recommender Systems [arXiv]
• Online and Offline Handwritten Chinese Character Recognition [arXiv]
• Tutorial on Variational Autoencoders [arXiv]
• Concrete Problems in AI Safety [arXiv]
• Deep Reinforcement Learning Discovers Internal Models [arXiv]
• SQuAD: 100,000+ Questions for Machine Comprehension of Text [arXiv]
• Conditional Image Generation with PixelCNN Decoders [arXiv]
• Model-Free Episodic Control [arXiv]
• Progressive Neural Networks [arXiv]
• Improved Techniques for Training GANs [arXiv] [code]
• Memory-Efficient Backpropagation Through Time [arXiv]
• InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets [arXiv]
• Zero-Resource Translation with Multi-Lingual Neural Machine Translation [arXiv]
• Key-Value Memory Networks for Directly Reading Documents [arXiv]
• Deep Recurrent Models with Fast-Forward Connections for Neural Machine Translatin [arXiv]
• Learning to learn by gradient descent by gradient descent [arXiv]
• Learning Language Games through Interaction [arXiv]
• Zoneout: Regularizing RNNs by Randomly Preserving Hidden Activations [arXiv]
• Smart Reply: Automated Response Suggestion for Email [arXiv]
• Virtual Adversarial Training for Semi-Supervised Text Classification [arXiv]
• Deep Reinforcement Learning for Dialogue Generation [arXiv]
• Very Deep Convolutional Networks for Natural Language Processing [arXiv]
• Neural Net Models for Open-Domain Discourse Coherence [arXiv]
• Neural Architectures for Fine-grained Entity Type Classification [arXiv]
• Matching Networks for One Shot Learning [arXiv]
• Cooperative Inverse Reinforcement Learning [arXiv] [article]
• Gated-Attention Readers for Text Comprehension [arXiv]
• End-to-end LSTM-based dialog control optimized with supervised and reinforcement learning [arXiv]
• Iterative Alternating Neural Attention for Machine Reading [arXiv]
• Memory-enhanced Decoder for Neural Machine Translation [arXiv]
• Multiresolution Recurrent Neural Networks: An Application to Dialogue Response Generation [arXiv]
• Learning to Optimize [arXiv] [article]
• Natural Language Comprehension with the EpiReader [arXiv]
• Conversational Contextual Cues: The Case of Personalization and History for Response Ranking [arXiv]
• Adversarially Learned Inference [arXiv]
• OpenAI Gym [arXiv] [code]
• Neural Network Translation Models for Grammatical Error Correction [arXiv]

2016-05

• Hierarchical Memory Networks [arXiv]
• Deep API Learning [arXiv]
• Wide Residual Networks [arXiv]
• TensorFlow: A system for large-scale machine learning [arXiv]
• Learning Natural Language Inference using Bidirectional LSTM model and Inner-Attention [arXiv]
• Aspect Level Sentiment Classification with Deep Memory Network [arXiv]
• FractalNet: Ultra-Deep Neural Networks without Residuals [arXiv]
• Learning End-to-End Goal-Oriented Dialog [arXiv]
• One-shot Learning with Memory-Augmented Neural Networks [arXiv]
• Deep Learning without Poor Local Minima [arXiv]
• AVEC 2016 - Depression, Mood, and Emotion Recognition Workshop and Challenge [arXiv]
• Data Programming: Creating Large Training Sets, Quickly [arXiv]
• Deeply-Fused Nets [arXiv]
• Deep Portfolio Theory [arXiv]
• Unsupervised Learning for Physical Interaction through Video Prediction [arXiv]
• Movie Description [arXiv]

2016-04

• Higher Order Recurrent Neural Networks [arXiv]
• Joint Line Segmentation and Transcription for End-to-End Handwritten Paragraph Recognition [arXiv]
• Hierarchical Deep Reinforcement Learning: Integrating Temporal Abstraction and Intrinsic Motivation [arXiv]
• The IBM 2016 English Conversational Telephone Speech Recognition System [arXiv]
• Dialog-based Language Learning [arXiv]
• Multilingual Part-of-Speech Tagging with Bidirectional Long Short-Term Memory Models and Auxiliary Loss [arXiv]
• Sentence-Level Grammatical Error Identification as Sequence-to-Sequence Correction [arXiv]
• A Network-based End-to-End Trainable Task-oriented Dialogue System [arXiv]
• Visual Storytelling [arXiv]
• Improving the Robustness of Deep Neural Networks via Stability Training [arXiv]
• Bridging the Gaps Between Residual Learning, Recurrent Neural Networks and Visual Cortex [arXiv]
• Scan, Attend and Read: End-to-End Handwritten Paragraph Recognition with MDLSTM Attention [arXiv]
• Sentence Level Recurrent Topic Model: Letting Topics Speak for Themselves [arXiv]
• Achieving Open Vocabulary Neural Machine Translation with Hybrid Word-Character Models [arXiv]
• Building Machines That Learn and Think Like People [arXiv]
• A Semisupervised Approach for Language Identification based on Ladder Networks [arXiv]
• Deep Networks with Stochastic Depth [arXiv]
• PHOCNet: A Deep Convolutional Neural Network for Word Spotting in Handwritten Documents [arXiv]

2016-03

• Improving Information Extraction by Acquiring External Evidence with Reinforcement Learning [arXiv]
• A Fast Unified Model for Parsing and Sentence Understanding [arXiv]
• Latent Predictor Networks for Code Generation [arXiv]
• Attend, Infer, Repeat: Fast Scene Understanding with Generative Models [arXiv]
• Recurrent Batch Normalization [arXiv]
• Neural Language Correction with Character-Based Attention [arXiv]
• Incorporating Copying Mechanism in Sequence-to-Sequence Learning [arXiv]
• How NOT To Evaluate Your Dialogue System [arXiv]
• Adaptive Computation Time for Recurrent Neural Networks [arXiv]
• A guide to convolution arithmetic for deep learning [arXiv]
• Colorful Image Colorization [arXiv]
• Unsupervised Learning of Visual Representations by Solving Jigsaw Puzzles [arXiv]
• Generating Factoid Questions With Recurrent Neural Networks: The 30M Factoid Question-Answer Corpus [arXiv]
• A Persona-Based Neural Conversation Model [arXiv]
• A Character-level Decoder without Explicit Segmentation for Neural Machine Translation [arXiv]
• Multi-Task Cross-Lingual Sequence Tagging from Scratch [arXiv]
• Neural Variational Inference for Text Processing [arXiv]
• Recurrent Dropout without Memory Loss [arXiv]
• One-Shot Generalization in Deep Generative Models [arXiv]
• Recursive Recurrent Nets with Attention Modeling for OCR in the Wild [[arXiv](Recursive Recurrent Nets with Attention Modeling for OCR in the Wild)]
• A New Method to Visualize Deep Neural Networks [[arXiv](A New Method to Visualize Deep Neural Networks)]
• Neural Architectures for Named Entity Recognition [arXiv]
• End-to-end Sequence Labeling via Bi-directional LSTM-CNNs-CRF [arXiv]
• Character-based Neural Machine Translation [arXiv]
• Learning Word Segmentation Representations to Improve Named Entity Recognition for Chinese Social Media [arXiv]

2016-02

• Architectural Complexity Measures of Recurrent Neural Networks [arXiv]
• Weight Normalization: A Simple Reparameterization to Accelerate Training of Deep Neural Networks [arXiv]
• Recurrent Neural Network Grammars [arXiv]
• Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations [arXiv]
• Contextual LSTM (CLSTM) models for Large scale NLP tasks [arXiv]
• Sequence-to-Sequence RNNs for Text Summarization [arXiv]
• Extraction of Salient Sentences from Labelled Documents [arXiv]
• Learning Distributed Representations of Sentences from Unlabelled Data [arXiv]
• Benefits of depth in neural networks [arXiv]
• Associative Long Short-Term Memory [arXiv]
• Why Should I Trust You?": Explaining the Predictions of Any Classifier [arXiv] [code]
• Generating images with recurrent adversarial networks [arXiv]
• Exploring the Limits of Language Modeling [arXiv]
• Swivel: Improving Embeddings by Noticing What’s Missing [arXiv]
• WebNav: A New Large-Scale Task for Natural Language based Sequential Decision Making [arXiv]
• Efficient Character-level Document Classification by Combining Convolution and Recurrent Layers [arXiv]
• Gradient Descent Converges to Minimizers [arXiv] [article]
• BinaryNet: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1 [arXiv]
• Learning Discriminative Features via Label Consistent Neural Network [arXiv]

2016-01

• What’s your ML test score? A rubric for ML production systems [Research at Google]
• Pixel Recurrent Neural Networks [arXiv]
• Bitwise Neural Networks [arXiv]
• Long Short-Term Memory-Networks for Machine Reading [arXiv]
• Coverage-based Neural Machine Translation [arXiv]
• Understanding Deep Convolutional Networks [arXiv]
• Training Recurrent Neural Networks by Diffusion [arXiv]
• Automatic Description Generation from Images: A Survey of Models, Datasets, and Evaluation Measures [arXiv]
• Multi-Way, Multilingual Neural Machine Translation with a Shared Attention Mechanism [arXiv]
• Recurrent Memory Network for Language Modeling [arXiv]
• Language to Logical Form with Neural Attention [arXiv]
• Learning to Compose Neural Networks for Question Answering [arXiv]
• The Inevitability of Probability: Probabilistic Inference in Generic Neural Networks Trained with Non-Probabilistic Feedback [arXiv]
• COCO-Text: Dataset and Benchmark for Text Detection and Recognition in Natural Images [arXiv]
• Survey on the attention based RNN model and its applications in computer vision [arXiv]

2015-12

NLP

• Strategies for Training Large Vocabulary Neural Language Models [arXiv]
• Multilingual Language Processing From Bytes [arXiv]
• Learning Document Embeddings by Predicting N-grams for Sentiment Classification of Long Movie Reviews [arXiv]
• Target-Dependent Sentiment Classification with Long Short Term Memory [arXiv]
• Reading Text in the Wild with Convolutional Neural Networks [arXiv]

Vision

• Deep Residual Learning for Image Recognition [arXiv]
• Rethinking the Inception Architecture for Computer Vision [arXiv]
• Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks [arXiv]
• Deep Speech 2: End-to-End Speech Recognition in English and Mandarin [arXiv]

2015-11

NLP

• Deep Reinforcement Learning with a Natural Language Action Space [arXiv]
• Sequence Level Training with Recurrent Neural Networks [arXiv]
• Teaching Machines to Read and Comprehend [arxiv]
• Semi-supervised Sequence Learning [arXiv]
• Multi-task Sequence to Sequence Learning [arXiv]
• Alternative structures for character-level RNNs [arXiv]
• Larger-Context Language Modeling [arXiv]
• A Unified Tagging Solution: Bidirectional LSTM Recurrent Neural Network with Word Embedding [arXiv]
• Towards Universal Paraphrastic Sentence Embeddings [arXiv]
• BlackOut: Speeding up Recurrent Neural Network Language Models With Very Large Vocabularies [arXiv]
• Sequence Level Training with Recurrent Neural Networks [arXiv]
• Natural Language Understanding with Distributed Representation [arXiv]
• sense2vec - A Fast and Accurate Method for Word Sense Disambiguation In Neural Word Embeddings [arXiv]
• LSTM-based Deep Learning Models for non-factoid answer selection [arXiv]

Programs

• Neural Random-Access Machines [arxiv]
• Neural Programmer: Inducing Latent Programs with Gradient Descent [arXiv]
• Neural Programmer-Interpreters [arXiv]
• Learning Simple Algorithms from Examples [arXiv]
• Neural GPUs Learn Algorithms [arXiv] [code]
• On Learning to Think: Algorithmic Information Theory for Novel Combinations of Reinforcement Learning Controllers and Recurrent Neural World Models [arXiv]

Vision

• ReSeg: A Recurrent Neural Network for Object Segmentation [arXiv]
• Deconstructing the Ladder Network Architecture [arXiv]
• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks [arXiv]
• Multi-Scale Context Aggregation by Dilated Convolutions [arXiv] [code]

General

• Towards Principled Unsupervised Learning [arXiv]
• Dynamic Capacity Networks [arXiv]
• Generating Sentences from a `ous Space [arXiv]
• Net2Net: Accelerating Learning via Knowledge Transfer [arXiv]
• A Roadmap towards Machine Intelligence [arXiv]
• Session-based Recommendations with Recurrent Neural Networks [arXiv]
• Regularizing RNNs by Stabilizing Activations [arXiv]

2015-10

• A Sensitivity Analysis of (and Practitioners' Guide to) Convolutional Neural Networks for Sentence Classification [arXiv]
• Attention with Intention for a Neural Network Conversation Model [arXiv]
• Part-of-Speech Tagging with Bidirectional Long Short-Term Memory Recurrent Neural Network [arXiv]
• A Survey: Time Travel in Deep Learning Space: An Introduction to Deep Learning Models and How Deep Learning Models Evolved from the Initial Ideas [arXiv]
• A Primer on Neural Network Models for Natural Language Processing [arXiv]
• A Diversity-Promoting Objective Function for Neural Conversation Models [arXiv]

2015-09

• Character-level Convolutional Networks for Text Classification [arXiv]
• A Neural Attention Model for Abstractive Sentence Summarization [arXiv]
• Poker-CNN: A Pattern Learning Strategy for Making Draws and Bets in Poker Games [arXiv]

2015-08

• Neural Machine Translation of Rare Words with Subword Units [arXiv] [code]
• Listen, Attend and Spell [arxiv]
• Character-Aware Neural Language Models [arXiv]
• Improved Transition-Based Parsing by Modeling Characters instead of Words with LSTMs [arXiv]
• Finding Function in Form: Compositional Character Models for Open Vocabulary Word Representation [arXiv]
• Effective Approaches to Attention-based Neural Machine Translation [arXiv]

2015-07

• Building End-To-End Dialogue Systems Using Generative Hierarchical Neural Network Models [arXiv]
• Semi-Supervised Learning with Ladder Networks [arXiv]
• Document Embedding with Paragraph Vectors [arXiv]
• Training Very Deep Networks [arXiv]

2015-06

• Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning [arXiv]
• A Neural Network Approach to Context-Sensitive Generation of Conversational Responses [arXiv]
• Document Embedding with Paragraph Vectors [arXiv]
• A Neural Conversational Model [arXiv]
• Skip-Thought Vectors [arXiv]
• Pointer Networks [arXiv]
• Spatial Transformer Networks [arXiv]
• Tree-structured composition in neural networks without tree-structured architectures [arXiv]
• Visualizing and Understanding Neural Models in NLP [arXiv]
• Learning to Transduce with Unbounded Memory [arXiv]
• Ask Me Anything: Dynamic Memory Networks for Natural Language Processing [arXiv]
• Deep Knowledge Tracing [arXiv]

2015-05

• ReNet: A Recurrent Neural Network Based Alternative to Convolutional Networks [arXiv]
• Reinforcement Learning Neural Turing Machines [arXiv]

2015-04

• Correlational Neural Networks [arXiv]

2015-03

• Distilling the Knowledge in a Neural Network [arXiv]
• End-To-End Memory Networks [arXiv]
• Neural Responding Machine for Short-Text Conversation [arXiv]
• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift [arXiv]
• Escaping From Saddle Points --- Online Stochastic Gradient for Tensor Decomposition [arXiv] [[article](Escaping from Saddle Points)]

2015-02

• Human-level control through deep reinforcement learning [Nature] [code]
• Text Understanding from Scratch [arXiv]
• Show, Attend and Tell: Neural Image Caption Generation with Visual Attention [arXiv]

2015-01

• Hidden Technical Debt in Machine Learning Systems [NIPS]

2014-12

• Learning Longer Memory in Recurrent Neural Networks [arXiv]
• Neural Turing Machines [arxiv]
• Grammar as a Foreign Langauage [arXiv]
• On Using Very Large Target Vocabulary for Neural Machine Translation [arXiv]
• Effective Use of Word Order for Text Categorization with Convolutional Neural Networks [arXiv]
• Multiple Object Recognition with Visual Attention [arXiv]

2014-11

• The Loss Surfaces of Multilayer Networks [arXiv]

2014-10

• Learning to Execute [arXiv]

2014-09

• Sequence to Sequence Learning with Neural Networks [arXiv]
• Neural Machine Translation by Jointly Learning to Align and Translate [arxiv]
• On the Properties of Neural Machine Translation: Encoder-Decoder Approaches [arXiv]
• Recurrent Neural Network Regularization [arXiv]
• Very Deep Convolutional Networks for Large-Scale Image Recognition [arXiv]
• Going Deeper with Convolutions [arXiv]

2014-08

• Convolutional Neural Networks for Sentence Classification [arxiv]

2014-07

2014-06

• Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation [arXiv]
• Recurrent Models of Visual Attention [arXiv]
• Generative Adversarial Networks [arXiv]

2014-05

• Distributed Representations of Sentences and Documents [arXiv]

2014-04

• A Convolutional Neural Network for Modelling Sentences [arXiv]

2014-03

2014-02

2014-01

• Machine Learning: The High Interest Credit Card of Technical Debt [Research at Google]

2013

• Visualizing and Understanding Convolutional Networks [arXiv]
• DeViSE: A Deep Visual-Semantic Embedding Model [pub]
• Maxout Networks [arXiv]
• Exploiting Similarities among Languages for Machine Translation [arXiv]
• Efficient Estimation of Word Representations in Vector Space [arXiv]

2011

• Natural Language Processing (almost) from Scratch [arXiv]

deeplearning-papernotes's Issues

Hope updates are coming

Hi, Denny! I have been following your notes for a long time since I was a college students.
Hope I can see some new thoughts from you!

May you provide more understanding of how model2(Condition softmax on n-grams before character) works?

I have read the paper, and I can not understand how model2 works, you said you thought you understand what it does, so could you please give me more explanations on the cond. model? Thanks very much.

fyi fork w additions https://github.com/trampolinerocket/deep-learning-papers

Short Science port?

These notes would be well suited to shortscience, the goal on that site being to provide concise summaries of papers in markdown.

Add link to code used in the paper

Could it be useful to have also the code used in the paper, when provided by the authors, next to the paper link? What do you think about a format of this type for the link's variable name: webRepository_[framework/library|Language]_model/packageName?
(e.g. tf-seq2seq link variable will be: [GH_TF_Tf-Seq2Seq] )
Or if you prefer something in the KISS way: a simple code as variable name will be perfect.

Add papers

GitXiv links

Thanks for this useful summary page. Have you considered providing also links to the relevant GitXiv entries, when available?

If you are not familiar with GitXiv, have a look: http://www.gitxiv.com/

For example, here is the WaveNet entry: http://www.gitxiv.com/posts/W5ax2LzW9Ht3mtTk2/wavenet-a-generative-model-for-raw-audio

Thanks again!

Response to notes on ``Neural Machine Translation with Reconstruction''

Hi Denny,

Thank you for your interest on our reconstruction work. I am the first author of this paper, and try to answer your questions.

• I feel like "adequacy" is a somewhat strange description of what the authors try to optimize. Wouldn't "coverage" be more appropriate?

Adequacy and/or Fluency evaluations are regularly employed for assessing the quality of machine translation. Adequacy measures how much of the meaning expressed in the source is also expressed in the target translation. It is well known that NMT favors fluent but inadequate translations, which have not only the coverage problems (e.g., over-translation and under-translation) but also mis-translated (e.g., wrong sense and unusual usage) and spurious translation (i.e., translation segments without any reference in source) problems.

• In Table 1, why does BLEU score still decrease when length normalization is applied? The authors don't go into detail on this.

As shown in Table 1, likelihood with length normalization favors long translations, which may face over-translation problems.

• The training curves are a bit confusing/missing. I would've liked to see a standard training curve that shows the MLE objective loss and the finetuning with reconstruction objective side-by-side.

We try to show that the increase of translation performance is indeed due to the improvement of reconstruction over time. We care more about the improvement of translation performance, since it is the ultimate goal of NMT.

• The training procedure somewhat confusing. The say "We further train the model for 10 epochs" with reconstruction objective, byt then "we use a trained model at iteration 110k". I'm assuming they do early-stopping at 110k * 80 = 8.8M steps. Again, would've liked to see the loss curves for this, not just BLEU curves.

During the training, we validate the translation performance for every 10K iterations. We select models that yield best performances on the validation set for NMT models. This is a standard procedure to select a well trained model in NMT.

Again, we care more about the ultimate goal of NMT -- translation performance measured in BLEU scores.

• I would've liked to see model performance on more "standard" NMT datasets like EN-FR and EN-DE, etc.

We only test on the Chinese-English translation task, which uses the same data as in our previous work Modeling Coverage for Neural Machine Translation and Context Gates for Neural Machine Translation.

• Is there perhaps a smarter way to do reconstruction iteratively by looking at what's missing from the reconstructed output? Trainig with reconstructor with MLE has some of the same drawbacks as training standard enc-dec with MLE and teacher forcing.

Really good point! There should be a better way to model the reconstruction (e.g., focus only on the wrong part). We will study on this in the future.

MLE favors fluent but inadequate translations, thus is not a optimal metric for NMT. It is necessary to introduce a better objective, such as sentence-level BLEU (Shen et al., 2016), and an auxiliary objective of reconstruction (in this work) or coverage penalty (in GNMT paper).