# Awesome Papers: 2017-01-4

### A K-fold Method for Baseline Estimation in Policy Gradient Algorithms

*Nithyanand Kota, Abhishek Mishra, Sunil Srinivasa, Xi (Peter) Chen, Pieter Abbeel*

The high variance issue in unbiased policy-gradient methods such as VPG and REINFORCE is typically mitigated by adding a baseline. However, the baseline fitting itself suffers from the underfitting or the overfitting problem. In this paper, we develop a K-fold method for baseline estimation in policy gradient algorithms.The parameter K is the baseline estimation hyperparameter that can adjust the bias-variance trade-off in the baseline estimates. We demonstrate the usefulness of our approach via two state-of-the-art policy gradient algorithms on three MuJoCo locomotive control tasks.

### Stochastic Planning and Lifted Inference

*Roni Khardon and Scott Sanner*

Lifted probabilistic inference (Poole, 2003) and symbolic dynamic programming for lifted stochastic planning (Boutilier et al, 2001) were introduced around the same time as algorithmic efforts to use abstraction in stochastic systems.Over the years, these ideas evolved into two distinct lines of research, each supported by a rich literature. Lifted probabilistic inference focused on efficient arithmetic operations on template-based graphical models under a finite domain assumption while symbolic dynamic programming focused on supporting sequential decision-making in rich quantified logical action models and on open domain reasoning. Given their common motivation but different focal points, both lines of research have yielded highly complementary innovations. In this chapter, we aim to help close the gap between these two research areas by providing an overview of lifted stochastic planning from the perspective of probabilistic inference, showing strong connections to other chapters in this book. This also allows us to define Generalized Lifted Inference as a paradigm that unifies these areas and elucidates open problems for future research that can benefit both lifted inference and stochastic planning.

### Generating Focussed Molecule Libraries for Drug Discovery with Recurrent Neural Networks

*Marwin H.S. Segler, Thierry Kogej, Christian Tyrchan, Mark P. Waller*

In de novo drug design, computational strategies are used to generate novel molecules with good affinity to the desired biological target. In this work, we show that recurrent neural networks can be trained as generative models for molecular structures, similar to statistical language models in natural language processing. We demonstrate that the properties of the generated molecules correlate very well with the properties of the molecules used to train the model. In order to enrich libraries with molecules active towards a given biological target, we propose to fine-tune the model with small sets of molecules, which are known to be active against that target.Against Staphylococcus aureus, the model reproduced 14% of 6051 hold-out test molecules that medicinal chemists designed, whereas against Plasmodium falciparum (Malaria) it reproduced 28% of 1240 test molecules. When coupled with a scoring function, our model can perform the complete de novo drug design cycle to generate large sets of novel molecules for drug discovery.

### Deep Convolutional Neural Networks for Pairwise Causality

*Karamjit Singh, Garima Gupta, Lovekesh Vig, Gautam Shroff, and Puneet Agarwal*

Discovering causal models from observational and interventional data is an important first step preceding what-if analysis or counterfactual reasoning. As has been shown before, the direction of pairwise causal relations can, under certain conditions, be inferred from observational data via standard gradient-boosted classifiers (GBC) using carefully engineered statistical features. In this paper we apply deep convolutional neural networks (CNNs) to this problem by plotting attribute pairs as 2-D scatter plots that are fed to the CNN as images. We evaluate our approach on the ‘Cause- Effect Pairs’ NIPS 2013 Data Challenge. We observe that a weighted ensemble of CNN with the earlier GBC approach yields significant improvement. Further, we observe that when less training data is available, our approach performs better than the GBC based approach suggesting that CNN models pre-trained to determine the direction of pairwise causal direction could have wider applicability in causal discovery and enabling what-if or counterfactual analysis.

### Akid: A Library for Neural Network Research and Production from a Dataism Approach

*Shuai Li*

Neural networks are a revolutionary but immature technique that is fast evolving and heavily relies on data. To benefit from the newest development and newly available data, we want the gap between research and production as small as possibly. On the other hand, differing from traditional machine learning models, neural network is not just yet another statistic model, but a model for the natural processing engine — the brain. In this work, we describe a neural network library named {\texttt akid}. It provides higher level of abstraction for entities (abstracted as blocks) in nature upon the abstraction done on signals (abstracted as tensors) by Tensorflow, characterizing the dataism observation that all entities in nature processes input and emit out in some ways. It includes a full stack of software that provides abstraction to let researchers focus on research instead of implementation, while at the same time the developed program can also be put into production seamlessly in a distributed environment, and be production ready. At the top application stack, it provides out-of-box tools for neural network applications. Lower down, akid provides a programming paradigm that lets user easily build customized models. The distributed computing stack handles the concurrency and communication, thus letting models be trained or deployed to a single GPU, multiple GPUs, or a distributed environment without affecting how a model is specified in the programming paradigm stack. Lastly, the distributed deployment stack handles how the distributed computing is deployed, thus decoupling the research prototype environment with the actual production environment, and is able to dynamically allocate computing resources, so development (Devs) and operations (Ops) could be separated.

链接：http://akid.readthedocs.io/en/latest/

### Deep Recurrent Q-Learning for Partially Observable MDPs

*Matthew Hausknecht, Peter Stone*

Deep Reinforcement Learning has yielded proficient controllers for complex tasks. However, these controllers have limited memory and rely on being able to perceive the complete game screen at each decision point. To address these shortcomings, this article investigates the effects of adding recurrency to a Deep Q-Network (DQN) by replacing the first post-convolutional fully-connected layer with a recurrent LSTM. The resulting \textit{Deep Recurrent Q-Network} (DRQN), although capable of seeing only a single frame at each timestep, successfully integrates information through time and replicates DQN’s performance on standard Atari games and partially observed equivalents featuring flickering game screens. Additionally, when trained with partial observations and evaluated with incrementally more complete observations, DRQN’s performance scales as a function of observability. Conversely, when trained with full observations and evaluated with partial observations, DRQN’s performance degrades less than DQN’s. Thus, given the same length of history, recurrency is a viable alternative to stacking a history of frames in the DQN’s input layer and while recurrency confers no systematic advantage when learning to play the game, the recurrent net can better adapt at evaluation time if the quality of observations changes.

### Tuning Recurrent Neural Networks with Reinforcement Learning

*Natasha Jaques, Shixiang Gu, Richard E. Turner, Douglas Eck*

The approach of training sequence models using supervised learning and next-step prediction suffers from known failure modes. For example, it is notoriously difficult to ensure multi-step generated sequences have coherent global structure. We propose a novel sequence-learning approach in which we use a pre-trained Recurrent Neural Network (RNN) to supply part of the reward value in a Reinforcement Learning (RL) model. Thus, we can refine a sequence predictor by optimizing for some imposed reward functions, while maintaining good predictive properties learned from data. We propose efficient ways to solve this by augmenting deep Q-learning with a cross-entropy reward and deriving novel off-policy methods for RNNs from KL control. We explore the usefulness of our approach in the context of music generation. An LSTM is trained on a large corpus of songs to predict the next note in a musical sequence. This Note RNN is then refined using our method and rules of music theory. We show that by combining maximum likelihood (ML) and RL in this way, we can not only produce more pleasing melodies, but significantly reduce unwanted behaviors and failure modes of the RNN, while maintaining information learned from data.

### 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*

Count-based exploration algorithms are known to perform near-optimally when used in conjunction with tabular reinforcement learning (RL) methods for solving small discrete Markov decision processes (MDPs). It is generally thought that count-based methods cannot be applied in high-dimensional state spaces, since most states will only occur once. Recent deep RL exploration strategies are able to deal with high-dimensional continuous state spaces through complex heuristics, often relying on optimism in the face of uncertainty or intrinsic motivation. In this work, we describe a surprising finding: a simple generalization of the classic count-based approach can reach near state-of-the-art performance on various high-dimensional and/or continuous deep RL benchmarks. States are mapped to hash codes, which allows to count their occurrences with a hash table. These counts are then used to compute a reward bonus according to the classic count-based exploration theory. We find that simple hash functions can achieve surprisingly good results on many challenging tasks. Furthermore, we show that a domain-dependent learned hash code may further improve these results. Detailed analysis reveals important aspects of a good hash function: 1) having appropriate granularity and 2) encoding information relevant to solving the MDP. This exploration strategy achieves near state-of-the-art performance on both continuous control tasks and Atari 2600 games, hence providing a simple yet powerful baseline for solving MDPs that require considerable exploration.

### Deep Convolutional Neural Networks for Pairwise Causality

*Karamjit Singh, Garima Gupta, Lovekesh Vig, Gautam Shroff, Puneet Agarwal*

Discovering causal models from observational and interventional data is an important first step preceding what-if analysis or counterfactual reasoning. As has been shown before, the direction of pairwise causal relations can, under certain conditions, be inferred from observational data via standard gradient-boosted classifiers (GBC) using carefully engineered statistical features. In this paper we apply deep convolutional neural networks (CNNs) to this problem by plotting attribute pairs as 2-D scatter plots that are fed to the CNN as images. We evaluate our approach on the ‘Cause- Effect Pairs’ NIPS 2013 Data Challenge. We observe that a weighted ensemble of CNN with the earlier GBC approach yields significant improvement. Further, we observe that when less training data is available, our approach performs better than the GBC based approach suggesting that CNN models pre-trained to determine the direction of pairwise causal direction could have wider applicability in causal discovery and enabling what-if or counterfactual analysis.

### Deep Learning for Time-Series Analysis

*John Cristian Borges Gamboa*

In many real-world application, e.g., speech recognition or sleep stage classification, data are captured over the course of time, constituting a Time-Series. Time-Series often contain temporal dependencies that cause two otherwise identical points of time to belong to different classes or predict different behavior. This characteristic generally increases the difficulty of analysing them. Existing techniques often depended on hand-crafted features that were expensive to create and required expert knowledge of the field. With the advent of Deep Learning new models of unsupervised learning of features for Time-series analysis and forecast have been developed. Such new developments are the topic of this paper: a review of the main Deep Learning techniques is presented, and some applications on Time-Series analysis are summaried. The results make it clear that Deep Learning has a lot to contribute to the field.

### Online Learning with Regularized Kernel for One-class Classification

*Chandan Gautam, Aruna Tiwari, Sundaram Suresh and Kapil Ahuja*

This paper presents an online learning with regularized kernel based one-class extreme learning machine (ELM) classifier and is referred as online RK-OC-ELM. The baseline kernel hyperplane model considers whole data in a single chunk with regularized ELM approach for offline learning in case of one-class classification (OCC). Further, the basic hyper plane model is adapted in an online fashion from stream of training samples in this paper. Two frameworks viz., boundary and reconstruction are presented to detect the target class in online RKOC-ELM. Boundary framework based one-class classifier consists of single node output architecture and classifier endeavors to approximate all data to any real number. However, one-class classifier based on reconstruction framework is an autoencoder architecture, where output nodes are identical to input nodes and classifier endeavor to reconstruct input layer at the output layer. Both these frameworks employ regularized kernel ELM based online learning and consistency based model selection has been employed to select learning algorithm parameters. The performance of online RK-OC-ELM has been evaluated on standard benchmark datasets as well as on artificial datasets and the results are compared with existing state-of-the art one-class classifiers. The results indicate that the online learning one-class classifier is slightly better or same as batch learning based approaches. As, base classifier used for the proposed classifiers are based on the ELM, hence, proposed classifiers would also inherit the benefit of the base classifier i.e. it will perform faster computation compared to traditional autoencoder based one-class classifier.