A5090720315- Recommender Systems and Techniques
- Topic Modeling
- Reinforcement Learning in Robotics
- Advanced Graph Neural Networks
- Natural Language Processing Techniques
- Advanced Bandit Algorithms Research
- Multimodal Machine Learning Applications
- Consumer Market Behavior and Pricing
- Domain Adaptation and Few-Shot Learning
- Auction Theory and Applications
- Data Stream Mining Techniques
- Generative Adversarial Networks and Image Synthesis
- Adversarial Robustness in Machine Learning
- Text and Document Classification Technologies
- Complex Network Analysis Techniques
- Image Retrieval and Classification Techniques
- Information Retrieval and Search Behavior
- Advanced Neural Network Applications
- Online Learning and Analytics
- Machine Learning and Data Classification
- Advanced Image and Video Retrieval Techniques
- Mobile Crowdsensing and Crowdsourcing
- Caching and Content Delivery
- Human Mobility and Location-Based Analysis
- Web Data Mining and Analysis
Shanghai Jiao Tong University
2016-2025
NIHR Imperial Biomedical Research Centre
2024
Imperial College London
2024
Research Center for Eco-Environmental Sciences
2024
University of Chinese Academy of Sciences
2024
Wenzhou University
2024
First Hospital of Qinhuangdao
2022-2023
Shenzhen University Health Science Center
2022-2023
Beijing Normal University
2021-2023
Chinese People's Liberation Army
2023
As a new way of training generative models, Generative Adversarial Net (GAN) that uses discriminative model to guide the has enjoyed considerable success in generating real-valued data. However, it limitations when goal is for sequences discrete tokens. A major reason lies outputs from make difficult pass gradient update model. Also, can only assess complete sequence, while partially generated non-trivial balance its current score and future one once entire sequence been generated. In this...
Domain adaptation aims at generalizing a high-performance learner on target domain via utilizing the knowledge distilled from source which has different but related data distribution. One solution to is learn invariant feature representations while learned should also be discriminative in prediction. To such representations, frameworks usually include representation learning approach measure and reduce discrepancy, as well discriminator for classification. Inspired by Wasserstein GAN, this...
Predicting user responses, such as clicks and conversions, is of great importance has found its usage inmany Web applications including recommender systems, websearch online advertising. The data in those applicationsis mostly categorical contains multiple fields, a typicalrepresentation to transform it into high-dimensional sparsebinary feature representation via one-hot encoding. Facing withthe extreme sparsity, traditional models may limit their capacityof mining shallow patterns from the...
This paper provides a unified account of two schools thinking in information retrieval modelling: the generative focusing on predicting relevant documents given query, and discriminative relevancy query-document pair. We propose game theoretical minimax to iteratively optimise both models. On one hand, model, aiming mine signals from labelled unlabelled data, guidance train model towards fitting underlying relevance distribution over query. other acting as an attacker current generates...
The goal of graph representation learning is to embed each vertex in a into low-dimensional vector space. Existing methods can be classified two categories: generative models that learn the underlying connectivity distribution graph, and discriminative predict probability edge existence between pair vertices. In this paper, we propose GraphGAN, an innovative framework unifying above classes methods, which model play game-theoretical minimax game. Specifically, for given vertex, tries fit its...
Automatically generating coherent and semantically meaningful text has many applications in machine translation, dialogue systems, image captioning, etc. Recently, by combining with policy gradient, Generative Adversarial Nets(GAN) that use a discriminative model to guide the training of generative as reinforcement learning shown promising results generation. However, scalar guiding signal is only available after entire been generated lacks intermediate information about structure during...
Techniques for automatically designing deep neural network architectures such as reinforcement learning based approaches have recently shown promising results. However, their success is on vast computational resources (e.g. hundreds of GPUs), making them difficult to be widely used. A noticeable limitation that they still design and train each from scratch during the exploration architecture space, which highly inefficient. In this paper, we propose a new framework toward efficient search by...
We introduce Texygen, a benchmarking platform to support research on open-domain text generation models. Texygen has not only implemented majority of models, but also covered set metrics that evaluate the diversity, quality and consistency generated texts. The could help standardize improve reproductivity reliability future work in generation.
Techniques for automatically designing deep neural network architectures such as reinforcement learning based approaches have recently shown promising results. However, their success is on vast computational resources (e.g. hundreds of GPUs), making them difficult to be widely used. A noticeable limitation that they still design and train each from scratch during the exploration architecture space, which highly inefficient. In this paper, we propose a new framework toward efficient search by...
Domain adaptation aims at generalizing a high-performance learner on target domain via utilizing the knowledge distilled from source which has different but related data distribution. One solution to is learn invariant feature representations while learned should also be discriminative in prediction. To such representations, frameworks usually include representation learning approach measure and reduce discrepancy, as well discriminator for classification. Inspired by Wasserstein GAN, this...
In this paper we study bid optimisation for real-time bidding (RTB) based display advertising. RTB allows advertisers to on a ad impression in real time when it is being generated. It goes beyond contextual advertising by motivating the focused user data and different from sponsored search auction where price associated with keywords. For demand side, fundamental technical challenge automate process budget, campaign objective various information gathered runtime history. paper, programmatic...
Existing multi-agent reinforcement learning methods are limited typically to a small number of agents. When the agent increases largely, becomes intractable due curse dimensionality and exponential growth interactions. In this paper, we present \emph{Mean Field Reinforcement Learning} where interactions within population agents approximated by those between single average effect from overall or neighboring agents; interplay two entities is mutually reinforced: individual agent's optimal...
Efficient H 2 O splitting for evolution over the semiconductor photocatalyst is a crucial strategy in field of energy and environment. Herein, cocatalyst‐free 2D–2D CdS/g‐C 3 N 4 step‐scheme (S‐scheme) heterojunction photocatalysts are fabricated through situ hydrothermal growth 2D CdS nanosheets (NSs) on g‐C NSs. The results clearly confirm that binary CdS/0.7g‐C S‐scheme shows best production rate (15.3 mmol g −1 h ) without using any cocatalyst, which 3.83 times 3060 higher than those...
Collaborative filtering techniques rely on aggregated user preference data to make personalized predictions. In many cases, users are reluctant explicitly express their preferences and recommender systems have infer them from implicit behaviors, such as clicking a link in webpage or playing music track. The clicks the plays good for indicating items liked (i.e., positive training examples), but did not like (negative examples) directly observed. Previous approaches either randomly pick...
User response prediction is a crucial component for personalized information retrieval and filtering scenarios, such as recommender system web search. The data in user mostly multi-field categorical format transformed into sparse representations via one-hot encoding. Due to the sparsity problems representation optimization, most research focuses on feature engineering shallow modeling. Recently, deep neural networks have attracted attention problem their high capacity end-to-end training...
Text-based question answering (TBQA) has been studied extensively in recent years. Most existing approaches focus on finding the answer to a within single paragraph. However, many difficult questions require multiple supporting evidence from scattered text among two or more documents. In this paper, we propose Dynamically Fused Graph Network (DFGN), novel method those requiring and reasoning over them. Inspired by human’s step-by-step behavior, DFGN includes dynamic fusion layer that starts...
We introduce Texygen, a benchmarking platform to support research on open-domain text generation models. Texygen has not only implemented majority of models, but also covered set metrics that evaluate the diversity, quality and consistency generated texts. The could help standardize facilitate sharing fine-tuned open-source implementations among researchers for their work. As consequence, this would in improving reproductivity reliability future work generation.
Molecular graph generation is a fundamental problem for drug discovery and has been attracting growing attention. The challenging since it requires not only generating chemically valid molecular structures but also optimizing their chemical properties in the meantime. Inspired by recent progress deep generative models, this paper we propose flow-based autoregressive model called GraphAF. GraphAF combines advantages of both approaches enjoys: (1) high flexibility data density estimation; (2)...
Learning feature interactions is crucial for click-through rate (CTR) prediction in recommender systems. In most existing deep learning models, are either manually designed or simply enumerated. However, enumerating all brings large memory and computation cost. Even worse, useless may introduce noise complicate the training process. this work, we propose a two-stage algorithm called Automatic Feature Interaction Selection (AutoFIS). AutoFIS can automatically identify important factorization...
The research field of Information Retrieval (IR) has evolved significantly, expanding beyond traditional search to meet diverse user information needs. Recently, Large Language Models (LLMs) have demonstrated exceptional capabilities in text understanding, generation, and knowledge inference, opening up exciting avenues for IR research. LLMs not only facilitate generative retrieval but also offer improved solutions model evaluation, user-system interactions. More importantly, the synergistic...
With the rapid development of online services and web applications, recommender systems (RS) have become increasingly indispensable for mitigating information overload matching users’ needs by providing personalized suggestions over items. Although RS research community has made remarkable progress past decades, conventional recommendation models (CRM) still some limitations, e.g. , lacking open-domain world knowledge, difficulties in comprehending underlying preferences motivations....