Genetic algorithms have been widely used in many practical optimization problems. Inspired by natural selection, operators, including mutation, crossover and selection, provide effective heuristics for search and black-box optimization. However, they have not been shown useful for deep reinforcement learning, possibly due to the catastrophic consequence of parameter crossovers of neural networks. Here, we present Genetic Policy Optimization (GPO), a new genetic algorithm for sample-efficient deep policy optimization. GPO uses imitation learning for policy crossover in the state space and applies policy gradient methods for mutation. Our experiments on MuJoCo tasks show that GPO as a genetic algorithm is able to provide superior performance over the state-of-the-art policy gradient methods and achieves comparable or higher sample efficiency.

Click to Read Paper
Hashing methods have been widely used for efficient similarity retrieval on large scale image database. Traditional hashing methods learn hash functions to generate binary codes from hand-crafted features, which achieve limited accuracy since the hand-crafted features cannot optimally represent the image content and preserve the semantic similarity. Recently, several deep hashing methods have shown better performance because the deep architectures generate more discriminative feature representations. However, these deep hashing methods are mainly designed for supervised scenarios, which only exploit the semantic similarity information, but ignore the underlying data structures. In this paper, we propose the semi-supervised deep hashing (SSDH) approach, to perform more effective hash function learning by simultaneously preserving semantic similarity and underlying data structures. The main contributions are as follows: (1) We propose a semi-supervised loss to jointly minimize the empirical error on labeled data, as well as the embedding error on both labeled and unlabeled data, which can preserve the semantic similarity and capture the meaningful neighbors on the underlying data structures for effective hashing. (2) A semi-supervised deep hashing network is designed to extensively exploit both labeled and unlabeled data, in which we propose an online graph construction method to benefit from the evolving deep features during training to better capture semantic neighbors. To the best of our knowledge, the proposed deep network is the first deep hashing method that can perform hash code learning and feature learning simultaneously in a semi-supervised fashion. Experimental results on 5 widely-used datasets show that our proposed approach outperforms the state-of-the-art hashing methods.

* 14 pages, accepted by IEEE Transactions on Circuits and Systems for Video Technology
Click to Read Paper
Hashing methods have attracted much attention for large scale image retrieval. Some deep hashing methods have achieved promising results by taking advantage of the strong representation power of deep networks recently. However, existing deep hashing methods treat all hash bits equally. On one hand, a large number of images share the same distance to a query image due to the discrete Hamming distance, which raises a critical issue of image retrieval where fine-grained rankings are very important. On the other hand, different hash bits actually contribute to the image retrieval differently, and treating them equally greatly affects the retrieval accuracy of image. To address the above two problems, we propose the query-adaptive deep weighted hashing (QaDWH) approach, which can perform fine-grained ranking for different queries by weighted Hamming distance. First, a novel deep hashing network is proposed to learn the hash codes and corresponding class-wise weights jointly, so that the learned weights can reflect the importance of different hash bits for different image classes. Second, a query-adaptive image retrieval method is proposed, which rapidly generates hash bit weights for different query images by fusing its semantic probability and the learned class-wise weights. Fine-grained image retrieval is then performed by the weighted Hamming distance, which can provide more accurate ranking than the traditional Hamming distance. Experiments on four widely used datasets show that the proposed approach outperforms eight state-of-the-art hashing methods.

* 13 pages, submitted to IEEE Transactions On Multimedia
Click to Read Paper
Deep reinforcement learning algorithms, including policy gradient methods and Q-learning, have been widely applied to a variety of decision-making problems. Their success has relied heavily on having very well designed dense reward signals, and therefore, they often perform badly on the sparse or episodic reward settings. Trajectory-based policy optimization methods, such as cross-entropy method and evolution strategies, do not take into consideration the temporal nature of the problem and often suffer from high sample complexity. Scaling up the efficiency of RL algorithms to real-world problems with sparse or episodic rewards is therefore a pressing need. In this work, we present a new perspective of policy optimization and introduce a self-imitation learning algorithm that exploits and explores well in the sparse and episodic reward settings. First, we view each policy as a state-action visitation distribution and formulate policy optimization as a divergence minimization problem. Then, we show that, with Jensen-Shannon divergence, this divergence minimization problem can be reduced into a policy-gradient algorithm with dense reward learned from experience replays. Experimental results indicate that our algorithm works comparable to existing algorithms in the dense reward setting, and significantly better in the sparse and episodic settings. To encourage exploration, we further apply the Stein variational policy gradient descent with the Jensen-Shannon kernel to learn multiple diverse policies and demonstrate its effectiveness on a number of challenging tasks.

Click to Read Paper
Recent advances in policy gradient methods and deep learning have demonstrated their applicability for complex reinforcement learning problems. However, the variance of the performance gradient estimates obtained from the simulation is often excessive, leading to poor sample efficiency. In this paper, we apply the stochastic variance reduced gradient descent (SVRG) to model-free policy gradient to significantly improve the sample-efficiency. The SVRG estimation is incorporated into a trust-region Newton conjugate gradient framework for the policy optimization. On several Mujoco tasks, our method achieves significantly better performance compared to the state-of-the-art model-free policy gradient methods in robotic continuous control such as trust region policy optimization (TRPO)

* 7 pages, 3 figures
Click to Read Paper
Deep hashing methods have received much attention recently, which achieve promising results by taking advantage of the strong representation power of deep networks. However, most existing deep hashing methods learn a whole set of hashing functions independently and directly, while ignore the correlation between different hashing functions that can promote the retrieval accuracy greatly. Inspired by the sequential decision ability of deep reinforcement learning, in this paper, we propose a new Deep Reinforcement Learning approach for Image Hashing (DRLIH). Our proposed DRLIH models the hashing learning problem as a Markov Decision Process (MDP), which learns each hashing function by correcting the errors imposed by previous ones and promotes retrieval accuracy. To the best of our knowledge, this is the first work that tries to address hashing problem from deep reinforcement learning perspective. The main contributions of our proposed DRLIH approach can be summarized as follows: (1) We propose a deep reinforcement learning hashing network. In our proposed DRLIH approach, we utilize recurrent neural network (RNN) as agents to model the hashing functions, which take actions of projecting images into binary codes sequentially, so that current hashing function learning can take previous hashing functions' error into account. (2) We propose a sequential learning strategy based on proposed DRLIH. We define the state as a tuple of internal features of RNN's hidden layers and image features, which can well reflect history decisions made by the agents. We also propose an action group method to enhance the correlation of the hash functions in the same group. Experiments on three widely-used datasets demonstrate the effectiveness of our proposed DRLIH approach.

* 18 pages, submitted to ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM). arXiv admin note: text overlap with arXiv:1612.02541
Click to Read Paper
Cross-modal hashing aims to map heterogeneous multimedia data into a common Hamming space, which can realize fast and flexible retrieval across different modalities. Supervised cross-modal hashing methods have achieved considerable progress by incorporating semantic side information. However, they mainly have two limitations: (1) Heavily rely on large-scale labeled cross-modal training data which are labor intensive and hard to obtain. (2) Ignore the rich information contained in the large amount of unlabeled data across different modalities, especially the margin examples that are easily to be incorrectly retrieved, which can help to model the correlations. To address these problems, in this paper we propose a novel Semi-supervised Cross-Modal Hashing approach by Generative Adversarial Network (SCH-GAN). We aim to take advantage of GAN's ability for modeling data distributions to promote cross-modal hashing learning in an adversarial way. The main contributions can be summarized as follows: (1) We propose a novel generative adversarial network for cross-modal hashing. In our proposed SCH-GAN, the generative model tries to select margin examples of one modality from unlabeled data when giving a query of another modality. While the discriminative model tries to distinguish the selected examples and true positive examples of the query. These two models play a minimax game so that the generative model can promote the hashing performance of discriminative model. (2) We propose a reinforcement learning based algorithm to drive the training of proposed SCH-GAN. The generative model takes the correlation score predicted by discriminative model as a reward, and tries to select the examples close to the margin to promote discriminative model by maximizing the margin between positive and negative data. Experiments on 3 widely-used datasets verify the effectiveness of our proposed approach.

* 12 pages, submitted to IEEE Transactions on Cybernetics
Click to Read Paper
Cross-modal hashing aims to map heterogeneous multimedia data into a common Hamming space, which can realize fast and flexible retrieval across different modalities. Unsupervised cross-modal hashing is more flexible and applicable than supervised methods, since no intensive labeling work is involved. However, existing unsupervised methods learn hashing functions by preserving inter and intra correlations, while ignoring the underlying manifold structure across different modalities, which is extremely helpful to capture meaningful nearest neighbors of different modalities for cross-modal retrieval. To address the above problem, in this paper we propose an Unsupervised Generative Adversarial Cross-modal Hashing approach (UGACH), which makes full use of GAN's ability for unsupervised representation learning to exploit the underlying manifold structure of cross-modal data. The main contributions can be summarized as follows: (1) We propose a generative adversarial network to model cross-modal hashing in an unsupervised fashion. In the proposed UGACH, given a data of one modality, the generative model tries to fit the distribution over the manifold structure, and select informative data of another modality to challenge the discriminative model. The discriminative model learns to distinguish the generated data and the true positive data sampled from correlation graph to achieve better retrieval accuracy. These two models are trained in an adversarial way to improve each other and promote hashing function learning. (2) We propose a correlation graph based approach to capture the underlying manifold structure across different modalities, so that data of different modalities but within the same manifold can have smaller Hamming distance and promote retrieval accuracy. Extensive experiments compared with 6 state-of-the-art methods verify the effectiveness of our proposed approach.

* 8 pages, accepted by 32th AAAI Conference on Artificial Intelligence (AAAI), 2018
Click to Read Paper
We propose a new localized inference algorithm for answering marginalization queries in large graphical models with the correlation decay property. Given a query variable and a large graphical model, we define a much smaller model in a local region around the query variable in the target model so that the marginal distribution of the query variable can be accurately approximated. We introduce two approximation error bounds based on the Dobrushin's comparison theorem and apply our bounds to derive a greedy expansion algorithm that efficiently guides the selection of neighbor nodes for localized inference. We verify our theoretical bounds on various datasets and demonstrate that our localized inference algorithm can provide fast and accurate approximation for large graphical models.

Click to Read Paper
We study the interpretability of conditional probability estimates for binary classification under the agnostic setting or scenario. Under the agnostic setting, conditional probability estimates do not necessarily reflect the true conditional probabilities. Instead, they have a certain calibration property: among all data points that the classifier has predicted P(Y = 1|X) = p, p portion of them actually have label Y = 1. For cost-sensitive decision problems, this calibration property provides adequate support for us to use Bayes Decision Theory. In this paper, we define a novel measure for the calibration property together with its empirical counterpart, and prove an uniform convergence result between them. This new measure enables us to formally justify the calibration property of conditional probability estimations, and provides new insights on the problem of estimating and calibrating conditional probabilities.

Click to Read Paper
Complex biological systems have been successfully modeled by biochemical and genetic interaction networks, typically gathered from high-throughput (HTP) data. These networks can be used to infer functional relationships between genes or proteins. Using the intuition that the topological role of a gene in a network relates to its biological function, local or diffusion based "guilt-by-association" and graph-theoretic methods have had success in inferring gene functions. Here we seek to improve function prediction by integrating diffusion-based methods with a novel dimensionality reduction technique to overcome the incomplete and noisy nature of network data. In this paper, we introduce diffusion component analysis (DCA), a framework that plugs in a diffusion model and learns a low-dimensional vector representation of each node to encode the topological properties of a network. As a proof of concept, we demonstrate DCA's substantial improvement over state-of-the-art diffusion-based approaches in predicting protein function from molecular interaction networks. Moreover, our DCA framework can integrate multiple networks from heterogeneous sources, consisting of genomic information, biochemical experiments and other resources, to even further improve function prediction. Yet another layer of performance gain is achieved by integrating the DCA framework with support vector machines that take our node vector representations as features. Overall, our DCA framework provides a novel representation of nodes in a network that can be used as a plug-in architecture to other machine learning algorithms to decipher topological properties of and obtain novel insights into interactomes.

* RECOMB 2015
Click to Read Paper
In this paper we present a new approach for tightening upper bounds on the partition function. Our upper bounds are based on fractional covering bounds on the entropy function, and result in a concave program to compute these bounds and a convex program to tighten them. To solve these programs effectively for general region graphs we utilize the entropy barrier method, thus decomposing the original programs by their dual programs and solve them with dual block optimization scheme. The entropy barrier method provides an elegant framework to generalize the message-passing scheme to high-order region graph, as well as to solve the block dual steps in closed-form. This is a key for computational relevancy for large problems with thousands of regions.

* Appears in Proceedings of the Twenty-Eighth Conference on Uncertainty in Artificial Intelligence (UAI2012)
Click to Read Paper
Deep neural networks have been remarkable successful in various AI tasks but often cast high computation and energy cost for energy-constrained applications such as mobile sensing. We address this problem by proposing a novel framework that optimizes the prediction accuracy and energy cost simultaneously, thus enabling effective cost-accuracy trade-off at test time. In our framework, each data instance is pushed into a cascade of deep neural networks with increasing sizes, and a selection module is used to sequentially determine when a sufficiently accurate classifier can be used for this data instance. The cascade of neural networks and the selection module are jointly trained in an end-to-end fashion by the REINFORCE algorithm to optimize a trade-off between the computational cost and the predictive accuracy. Our method is able to simultaneously improve the accuracy and efficiency by learning to assign easy instances to fast yet sufficiently accurate classifiers to save computation and energy cost, while assigning harder instances to deeper and more powerful classifiers to ensure satisfiable accuracy. With extensive experiments on several image classification datasets using cascaded ResNet classifiers, we demonstrate that our method outperforms the standard well-trained ResNets in accuracy but only requires less than 20% and 50% FLOPs cost on the CIFAR-10/100 datasets and 66% on the ImageNet dataset, respectively.

Click to Read Paper
Policy gradient methods have been successfully applied to many complex reinforcement learning problems. However, policy gradient methods suffer from high variance, slow convergence, and inefficient exploration. In this work, we introduce a maximum entropy policy optimization framework which explicitly encourages parameter exploration, and show that this framework can be reduced to a Bayesian inference problem. We then propose a novel Stein variational policy gradient method (SVPG) which combines existing policy gradient methods and a repulsive functional to generate a set of diverse but well-behaved policies. SVPG is robust to initialization and can easily be implemented in a parallel manner. On continuous control problems, we find that implementing SVPG on top of REINFORCE and advantage actor-critic algorithms improves both average return and data efficiency.

Click to Read Paper
In this paper, we propose a general approach to optimize anchor boxes for object detection. Nowadays, anchor boxes are widely adopted in state-of-the-art detection frameworks. However, all these frameworks pre-define anchor box shapes in a heuristic way and fix the size during training. To improve the accuracy and reduce the effort to design the anchor boxes, we propose to dynamically learn the shapes, which allows the anchors to automatically adapt to the data distribution and the network learning capability. The learning approach can be easily implemented in the stochastic gradient descent way and be plugged into any anchor box-based detection framework. The extra training cost is almost negligible and it has no impact on the inference time cost. Exhaustive experiments also demonstrate that the proposed anchor optimization method consistently achieves significant improvement ($\ge 1\%$ mAP absolute gain) over the baseline method on several benchmark datasets including Pascal VOC 07+12, MS COCO and Brainwash. Meanwhile, the robustness is also verified towards different anchor box initialization methods, which greatly simplifies the problem of anchor box design.

Click to Read Paper
Learning distributed representations for nodes in graphs is a crucial primitive in network analysis with a wide spectrum of applications. Linear graph embedding methods learn such representations by optimizing the likelihood of both positive and negative edges while constraining the dimension of the embedding vectors. We argue that the generalization performance of these methods is not due to the dimensionality constraint as commonly believed, but rather the small norm of embedding vectors. Both theoretical and empirical evidence are provided to support this argument: (a) we prove that the generalization error of these methods can be bounded by limiting the norm of vectors, regardless of the embedding dimension; (b) we show that the generalization performance of linear graph embedding methods is correlated with the norm of embedding vectors, which is small due to the early stopping of SGD and the vanishing gradients. We performed extensive experiments to validate our analysis and showcased the importance of proper norm regularization in practice.

Click to Read Paper
We propose a novel methodology to generate domain-specific large-scale question answering (QA) datasets by re-purposing existing annotations for other NLP tasks. We demonstrate an instance of this methodology in generating a large-scale QA dataset for electronic medical records by leveraging existing expert annotations on clinical notes for various NLP tasks from the community shared i2b2 datasets. The resulting corpus (emrQA) has 1 million question-logical form and 400,000+ question-answer evidence pairs. We characterize the dataset and explore its learning potential by training baseline models for question to logical form and question to answer mapping.

* Accepted at Conference on Empirical Methods in Natural Language Processing (EMNLP) 2018
Click to Read Paper
Representing data in hyperbolic space can effectively capture latent hierarchical relationships. With the goal of enabling accurate classification of points in hyperbolic space while respecting their hyperbolic geometry, we introduce hyperbolic SVM, a hyperbolic formulation of support vector machine classifiers, and elucidate through new theoretical work its connection to the Euclidean counterpart. We demonstrate the performance improvement of hyperbolic SVM for multi-class prediction tasks on real-world complex networks as well as simulated datasets. Our work allows analytic pipelines that take the inherent hyperbolic geometry of the data into account in an end-to-end fashion without resorting to ill-fitting tools developed for Euclidean space.

Click to Read Paper
The performance of off-policy learning, including deep Q-learning and deep deterministic policy gradient (DDPG), critically depends on the choice of the exploration policy. Existing exploration methods are mostly based on adding noise to the on-going actor policy and can only explore \emph{local} regions close to what the actor policy dictates. In this work, we develop a simple meta-policy gradient algorithm that allows us to adaptively learn the exploration policy in DDPG. Our algorithm allows us to train flexible exploration behaviors that are independent of the actor policy, yielding a \emph{global exploration} that significantly speeds up the learning process. With an extensive study, we show that our method significantly improves the sample-efficiency of DDPG on a variety of reinforcement learning tasks.

* 10 pages
Click to Read Paper
Protein secondary structure (SS) prediction is important for studying protein structure and function. When only the sequence (profile) information is used as input feature, currently the best predictors can obtain ~80% Q3 accuracy, which has not been improved in the past decade. Here we present DeepCNF (Deep Convolutional Neural Fields) for protein SS prediction. DeepCNF is a Deep Learning extension of Conditional Neural Fields (CNF), which is an integration of Conditional Random Fields (CRF) and shallow neural networks. DeepCNF can model not only complex sequence-structure relationship by a deep hierarchical architecture, but also interdependency between adjacent SS labels, so it is much more powerful than CNF. Experimental results show that DeepCNF can obtain ~84% Q3 accuracy, ~85% SOV score, and ~72% Q8 accuracy, respectively, on the CASP and CAMEO test proteins, greatly outperforming currently popular predictors. As a general framework, DeepCNF can be used to predict other protein structure properties such as contact number, disorder regions, and solvent accessibility.

Click to Read Paper