Search

Continual learning is the ability of agents to improve their capacities throughout multiple tasks continually. While recent works in the literature of continual learning mostly focused on developing either particular loss functions or specialized structures of neural network explaining the episodic memory or neural plasticity, we study continual learning from the perspective of the training mechanism. Specifically, we propose a COnitnual Match BAsed Training (COMBAT) framework for training a population of advantage-actor-critic (A2C) agents in Pommerman, a partially observable multi-agent environment with no communication. Following the COMBAT framework, we trained an agent, namely, Navocado, that won the title of the top 1 learning agent in the NeurIPS 2018 Pommerman Competition. Two critical features of our agent are worth mentioning. Firstly, our agent did not learn from any demonstrations. Secondly, our agent is highly reproducible. As a technical report, we articulate the design of state space, action space, reward, and most importantly, the COMBAT framework for our Pommerman agent. We show in the experiments that Pommerman is a perfect environment for studying continual learning, and the agent can improve its performance by continually learning new skills without forgetting the old ones. Finally, the result in the Pommerman Competition verifies the robustness of our agent when competing with various opponents., Comment: 8 pages, 7 figures

The effective of information retrieval (IR) systems have become more important than ever. Deep IR models have gained increasing attention for its ability to automatically learning features from raw text; thus, many deep IR models have been proposed recently. However, the learning process of these deep IR models resemble a black box. Therefore, it is necessary to identify the difference between automatically learned features by deep IR models and hand-crafted features used in traditional learning to rank approaches. Furthermore, it is valuable to investigate the differences between these deep IR models. This paper aims to conduct a deep investigation on deep IR models. Specifically, we conduct an extensive empirical study on two different datasets, including Robust and LETOR4.0. We first compared the automatically learned features and hand-crafted features on the respects of query term coverage, document length, embeddings and robustness. It reveals a number of disadvantages compared with hand-crafted features. Therefore, we establish guidelines for improving existing deep IR models. Furthermore, we compare two different categories of deep IR models, i.e. representation-focused models and interaction-focused models. It is shown that two types of deep IR models focus on different categories of words, including topic-related words and query-related words., SIGIR 2017 Workshop on Neural Information Retrieval (Neu-IR'17)

In recent years, deep neural models have been widely adopted for text matching tasks, such as question answering and information retrieval, showing improved performance as compared with previous methods. In this paper, we introduce the MatchZoo toolkit that aims to facilitate the designing, comparing and sharing of deep text matching models. Specifically, the toolkit provides a unified data preparation module for different text matching problems, a flexible layer-based model construction process, and a variety of training objectives and evaluation metrics. In addition, the toolkit has implemented two schools of representative deep text matching models, namely representation-focused models and interaction-focused models. Finally, users can easily modify existing models, create and share their own models for text matching in MatchZoo., 2 pages, 1 figures, Neu-IR: The SIGIR 2017 Workshop on Neural Information Retrieval

Deep neural networks have been successfully applied to many text matching tasks, such as paraphrase identification, question answering, and machine translation. Although ad-hoc retrieval can also be formalized as a text matching task, few deep models have been tested on it. In this paper, we study a state-of-the-art deep matching model, namely MatchPyramid, on the ad-hoc retrieval task. The MatchPyramid model employs a convolutional neural network over the interactions between query and document to produce the matching score. We conducted extensive experiments to study the impact of different pooling sizes, interaction functions and kernel sizes on the retrieval performance. Finally, we show that the MatchPyramid models can significantly outperform several recently introduced deep matching models on the retrieval task, but still cannot compete with the traditional retrieval models, such as BM25 and language models., Neu-IR '16 SIGIR Workshop on Neural Information Retrieval

Semantic matching, which aims to determine the matching degree between two texts, is a fundamental problem for many NLP applications. Recently, deep learning approach has been applied to this problem and significant improvements have been achieved. In this paper, we propose to view the generation of the global interaction between two texts as a recursive process: i.e. the interaction of two texts at each position is a composition of the interactions between their prefixes as well as the word level interaction at the current position. Based on this idea, we propose a novel deep architecture, namely Match-SRNN, to model the recursive matching structure. Firstly, a tensor is constructed to capture the word level interactions. Then a spatial RNN is applied to integrate the local interactions recursively, with importance determined by four types of gates. Finally, the matching score is calculated based on the global interaction. We show that, after degenerated to the exact matching scenario, Match-SRNN can approximate the dynamic programming process of longest common subsequence. Thus, there exists a clear interpretation for Match-SRNN. Our experiments on two semantic matching tasks showed the effectiveness of Match-SRNN, and its ability of visualizing the learned matching structure., Accepted by IJCAI-2016

Storm surge is one of the predominant natural threats to coastal communities. It occurs frequently and usually results in the most serious economic loss among all the natural hazards in China. The disaster induced by storm surge depends on various influencing factors such as the intensity, duration, and route of the passing typhoon, and thus a comprehensive understanding of natural coastal hazards is essential for the safe and sustainable development of the coastal area. Seawalls are the primary coastal protection facilities, and their protection standard can be established using return periods of the design tidal level (or water level) and design wave. But in China, the correlation between tidal level and wave, which is the parameter may enhance the interpretation of the actual storm process, isn't described in the existing design criterion. This paper proposes a new method to classify storm surge intensity based on PBGLD and PBLD. We illustrate this new method using observed tidal level and corresponding wave height series sampled from typhoon processes in the coastal area of Qingdao and Shenzhen, including the occurrence frequency of typhoon, and to make up the defects of the warning water level. A novel criterion is put forward to grade the intensity of typhoon-induced storm surge, and it forms the foundation to evaluate the storm surge intensity for seawall design. Study indicates that the new criterion gives an explicit notion in terms of joint probability that is easy to operate and fits well to the classification of the surge intensity. The proposed method of storm surge classification using new statistical model can be adopted as a reference approach in many coastal areas affected by typhoon. The data used here is limited, it is recommended that the calculation should only be used for reference, and more verification and improvement may be needed for broader application.

We have developed an incandescent Mo source to fabricate large-area single-crystalline MoSe2 thin films. The as-grown MoSe2 thin films were characterized using transmission electron microscopy, energy dispersive X-ray spectroscopy, atomic force microscopy, Raman spectroscopy, photoluminescence, reflection high energy electron diffraction (RHEED) and angular resolved photoemission spectroscopy (ARPES). A new Raman characteristic peak at 1591 cm-1 was identified. Results from Raman spectroscopy, photoluminescence, RHEED and ARPES studies consistently reveal that large-area single crystalline mono-layer of MoSe2 could be achieved by this technique. This technique enjoys several advantages over conventional approaches and could be extended to the growth of other two-dimensional layered materials containing a low-vapor-pressure element.

This paper will focus on the effect of atmospheric conditions on EO sensor performance using computer models. We have shown the importance of accurately modeling atmospheric effects for predicting the performance of an EO sensor. A simple example will demonstrated how real conditions for several sites in China will significantly impact on image correction, hyperspectral imaging, and remote sensing. The current state-of-the-art model for computing atmospheric transmission and radiance is, MODTRAN® 5, developed by the US Air Force Research Laboratory and Spectral Science, Inc. Research by the US Air Force, Navy and Army resulted in the public release of LOWTRAN 2 in the early 1970’s. Subsequent releases of LOWTRAN and MODTRAN® have continued until the present. Please verify that (1) all pages are present, (2) all figures are correct, (3) all fonts and special characters are correct, and (4) all text and figures fit within the red margin lines shown on this review document. Complete formatting information is available at http://SPIE.org/manuscripts Return to the Manage Active Submissions page at http://spie.org/submissions/tasks.aspx and approve or disapprove this submission. Your manuscript will not be published without this approval. Please contact author_help@spie.org with any questions or concerns. The paper will demonstrate the importance of using validated models and local measured meteorological, atmospheric and aerosol conditions to accurately simulate the atmospheric transmission and radiance. Frequently default conditions are used which can produce errors of as much as 75% in these values. This can have significant impact on remote sensing applications.

This paper describes the solution of Bazinga Team for Tmall Recommendation Prize 2014. With real-world user action data provided by Tmall, one of the largest B2C online retail platforms in China, this competition requires to predict future user purchases on Tmall website. Predictions are judged on F1Score, which considers both precision and recall for fair evaluation. The data set provided by Tmall contains more than half billion action records from over ten million distinct users. Such massive data volume poses a big challenge, and drives competitors to write every single program in MapReduce fashion and run it on distributed cluster. We model the purchase prediction problem as standard machine learning problem, and mainly employ regression and classification methods as single models. Individual models are then aggregated in a two-stage approach, using linear regression for blending, and finally a linear ensemble of blended models. The competition is approaching the end but still in running during writing this paper. In the end, our team achieves F1Score 6.11 and ranks 7th (out of 7,276 teams in total)., Accepted by 2nd Large Scale Recommender Systems Workshop, RecSys 2014