Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Khanh VuSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works86 from public data
- Domain-Specific Language Model Pretraining for Biomedical Natural Language Processing2,830
It is shown that for domains with abundant unlabeled text, such as biomedicine, pretraining language models from scratch results in substantial gains over continual pretraining of general-domain language models.
- Language Models for Image Captioning: The Quirks and What Works293
By combining key aspects of the ME and RNN methods, this paper achieves a new record performance over previously published results on the benchmark COCO dataset, however, the gains the authors see in BLEU do not translate to human judgments.
- Augmenting Language Models with Long-Term Memory263
Experiments show that the proposed framework, Language Models Augmented with Long-Term Memory (LongMem), outperforms strong long-context models on ChapterBreak, a challenging long- context modeling benchmark, and achieves remarkable improvements on memory-augmented in-context learning over LLMs.
- Fine-tuning large neural language models for biomedical natural language processing203
Overall, domain-specific vocabulary and pretraining facilitate robust models for fine-tuning and establish a new state of the art on a wide range of biomedical NLP applications.
- Dialogue State Tracking with a Language Model using Schema-Driven Prompting151
A new variation of the language modeling approach that uses schema-driven prompting to provide task-aware history encoding that is used for both categorical and non-categorical slots is introduced.
- 110
- 93
- 89
- Self-Verification Improves Few-Shot Clinical Information Extraction85
This work explores a general mitigation framework using self-verification, which leverages the LLM to provide provenance for its own extraction and check its own outputs and consistently improves accuracy for various LLMs in standard clinical information extraction tasks.
- Constrained locally weighted clustering75
The theoretical analysis shows that the probability of points being assigned to the correct clusters is much higher by the new algorithm, compared to the conventional methods, indicating that the design produces clusters which are closer to the ground truth than clusters created by the current state of theart algorithms.
- ExACT: Teaching AI Agents to Explore with Reflective-MCTS and Exploratory Learning67
ExACT, an approach to combine test-time search and self-learning to build o1-like models for agentic applications, is presented, suggesting a promising research direction to enhance VLMs' capabilities for agentic applications via test-time search and self-learning.
- UnitedQA: A Hybrid Approach for Open Domain Question Answering67
It is demonstrated that a simple hybrid approach by combining answers from both readers can efficiently take advantages of extractive and generative answer inference strategies and outperforms single models as well as homogeneous ensembles.
- 64
- The Microsoft Toolkit of Multi-Task Deep Neural Networks for Natural Language Understanding63
MT-DNN is presented, an open-source natural language understanding (NLU) toolkit that makes it easy for researchers and developers to train customized deep learning models, designed to facilitate rapid customization for a broad spectrum of NLU tasks.
- Sounding Board: A User-Centric and Content-Driven Social Chatbot61
The system architecture consists of several components including spoken language processing, dialogue management, language generation, and content management, with emphasis on user-centric and content-driven design.
- Disentangled Feature Representation for Few-Shot Image Classification57
This work proposes a novel disentangled feature representation (DFR) framework, dubbed DFR, for few-shot learning applications, which can adaptively decouple the discriminative features that are modeled by the classification branch, from the class-irrelevant component of the variation branch.
- 57
- Open-Domain Name Error Detection using a Multi-Task RNN48
A multi-task recurrent neural network language model for sentence-level name detection is proposed for use in combination with out-of-vocabulary word detection, which shows a 26% improvement in name-error detection F-score over a system using n-gram lexical features.
- Bi-directional Attention with Agreement for Dependency Parsing47
A novel bi-directional attention model for dependency parsing, which learns to agree on headword predictions from the forward and backward parsing directions, which achieves state-of-the-art unlabeled attachment scores on 6 languages.
- 36
- A non-linear dimensionality-reduction technique for fast similarity search in large databases36
A non-linear transformation scheme that extracts two important parameters specifying the data is presented, and it is proved that these parameters correspond to a bounded volume around the search sphere, irrespective of dimensionality.
- Graph Neural Networks With Triple Attention for Few-Shot Learning33
The proposed Attentive GNN model outperforms the state-of-the-art few-shot learning methods using both GNN and non-GNN approaches and is consistent over the mini-Image net, tiered-ImageNet, CUB-200-2011, and Flowers-102 benchmarks.
- Posterior Differential Regularization with f-divergence for Improving Model Robustness33
It is shown that regularizing the posterior difference with f-divergence can result in well-improved model robustness, indicating the great potential of the proposed framework for enhancing NLPmodel robustness.
- Reassessing Layer Pruning in LLMs: New Insights and Methods30
The results demonstrate that a simple approach, i.e., pruning the final 25\% of layers followed by fine-tuning the remaining last three layer, yields remarkably strong performance.
- Scalable and Sound Low-Rank Tensor Learning28
This work proposes directly optimizing the tensor trace norm by approximating its dual spectral norm, and shows that the approximation bounds can be efficiently converted to the original problem via the generalized conditional gradient algorithm.
- OrchestraLLM: Efficient Orchestration of Language Models for Dialogue State Tracking23
A novel SLM/LLM routing framework designed to improve computational efficiency and enhance task performance in dialogue state tracking tasks and reduces the computational costs by over 50%.
- Probabilistic Assumptions Matter: Improved Models for Distantly-Supervised Document-Level Question Answering22
This work addresses the problem of extractive question answering using document-level distant super-vision, pairing questions and relevant documents with answer strings and demonstrates that a multi-objective model can efficiently combine the advantages of multiple assumptions and outperform the best individual formulation.
- A Dynamic Speaker Model for Conversational Interactions22
A neural model for learning a dynamically updated speaker embedding in a conversational context is introduced and is shown to be useful for content ranking in a socialbot and dialog act prediction in human-human conversations.
- Site Selection via Learning Graph Convolutional Neural Networks: A Case Study of Singapore20
This work proposes to learn a graph convolutional network (GCN) for highly effective site selection tasks and presents a novel dataset that encompasses land use information as well as public transport networks in Singapore as a case study to benchmark site selection algorithms.
- Attentive Graph Neural Networks for Few-Shot Learning20
This work proposes a novel Attentive GNN (AGNN) to tackle few-shot learning challenges by incorporating a triple-attention mechanism, i.e., node self-att attention, neighborhood attention, and layer memory attention.
- Domain-Specific Pretraining for Vertical Search: Case Study on Biomedical Literature17
This work proposes a general approach for vertical search based on domain-specific pretraining and presents a case study for the biomedical domain, which performs comparably or better than the best systems in the official TREC-COVID evaluation, a COVID-related biomedical search competition.
- A Factored Neural Network Model for Characterizing Online Discussions in Vector Space17
A novel factored neural model that learns comment embeddings in an unsupervised way leveraging the structure of distributional context in online discussion forums and captures community style and topic, as well as response trigger patterns.
- SubSpace Projection: A unified framework for a class of partition-based dimension reduction techniques17
This work designs a greedy algorithm to efficiently determine a good partitioning of the data dimensions and proposes SubSpace Projection (SSP) as a unified framework for these partition-based techniques.
- Learning Latent Local Conversation Modes for Predicting Comment Endorsement in Online Discussions16
This paper addresses the problem of predicting community endorsement in online discussions, leveraging both the participant response structure and the text of the comment, using a novel architecture to learn latent modes of discussion structure that perform as well as deep neural networks but are more interpretable.
- Characterizing the Representer Theorem16
The necessary and sufficient conditions on the regularizer that ensure the representer theorem holds are characterized and the results are surprisingly simple yet broaden the conditions where the representser theorem is known to hold.
- 15
- Dyna-Think: Synergizing Reasoning, Acting, and World Model Simulation in AI Agents13
Dyna-Think, a thinking framework that integrates planning with an internal world model with reasoning and acting to enhance AI agent performance, is proposed and a promising research direction to integrate world model simulation into AI agents to enhance their reasoning, planning, and acting capabilities is suggested.
- 13
- A Multi-Directional Search technique for image annotation propagation13
A novel Multi-Directional Search framework for semi-automatic annotation propagation that provides accurate annotation assistance to the user - images with the same semantic meaning but different visual characteristics can be handled effectively.
- Targeted Adversarial Training for Natural Language Understanding12
Experiments show that TAT can significantly improve accuracy over standard adversarial training on GLUE and attain new state-of-the-art zero-shot results on XNLI.
- 11
- 10
- 10
- STSP: Spatial-Temporal Subspace Projection for Video Class-Incremental Learning9
This work proposes a discriminative Temporal-based Subspace Classifier (TSC) that represents each class with an orthogonal subspace basis and adopts subspace projection loss for classification, and implements inter-and intra-class orthogonal constraints into TSC.
- Train Once, and Decode As You Like9
A unified approach for supporting different generation manners of machine translation, including autoregressive, semi-autorgressive, and refinement-based non-autoregressive models, which achieves better or competitive translation performance compared with some strong task-specific baseline models in all the settings.
- Leveraging user query log: toward improving image data clustering9
This paper proposes a semi-supervised clustering algorithm, which leverages the accumulated user query log to guide the clustering process and can produce image clusters close to the ground truth and satisfying the semantic relations between the images inferred from the query log.
- 9
- Dyna-Mind: Learning to Simulate from Experience for Better AI Agents8
Dyna-Mind, a two-stage training framework that explicitly teaches (V)LM agents to integrate simulation ability into AI agents, and Dyna-GRPO, an online reinforcement learning method that leverages outcome and interaction-level signals to learn better policies for long-horizon, planning-intensive tasks are introduced.
- An automatic feature generation approach to multiple instance learning and its applications to image databases8
A novel learning technique is proposed which transforms the MIL problem into a standard supervised learning problem by defining a feature vector for each image bag and demonstrates the superiority of the proposed technique in categorization accuracy as compared to state-of-the-art methods.
- Semi-supervised dimensionality reduction in image feature space8
A semi-supervised reduction method that leverages relevance feedback information in the retrieval process to learn suitable linear and orthogonal embeddings is proposed.
- Second Rethinking of Network Pruning in the Adversarial Setting.7
A framework of concurrent adversarialTraining and weight pruning that enables model compression while still preserving the adversarial robustness and essentially tackles the dilemma of adversarial training is proposed.
- 5
- 5
- 5
- 5
- Lattice-Aided Extraction of Spread-Spectrum Hidden Data4
Two novel SS watermarking extraction approaches are proposed by integrating precise lattice decoding algorithms, resulting in a new method termed multi-carrier iterative successive interference cancellation (M-ISIC) and the near-optimal sphere decoding (SD) technique for referential extraction in SS watermarking.
- Transfer Attack for Bad and Good: Explain and Boost Adversarial Transferability across Multimodal Large Language Models4
This paper proposes a boosting method called Typography Augment Transfer-ability Method (TATM) to investigate the adversarial transferability performance across MLLMs further and demonstrates exceptional performance in real-world applications of "Harmful Word Insertion" and "Important Information Protection."
- DocLens: Multi-aspect Fine-grained Evaluation for Medical Text Generation4
A comprehensive human evaluation confirms that the proposed GPT-4-based metrics exhibit substantially higher agreement with human judgments than existing evaluation metrics, contributing to the understanding of medical text generation evaluation and offers a more reliable alternative to existing metrics.
- ACT-Diffusion: Efficient Adversarial Consistency Training for One-step Diffusion Models4
This paper proposes Adversarial Consistency Training (ACT), which directly minimizes the Jensen-Shannon divergence be-tween distributions at each timestep using a discriminator, and theoretically, ACT enhances generation quality, and convergence.
- 4
- 4
- 3
- 3
- 3
- Image retrieval in multipoint queries3
The results indicate that the new measure captures semantic image classes very well, and the superiority of the approach over the recent techniques is evident in simulated and realistic environments.
- 2
- 2
- 2
- 1
- 1
- Dynamical Pattern Classification of Lorenz System and Chen System1
This paper investigates the construction of the recognition system for classification of Lorenz system and Chen system, both of which can generate various types of dynamical patterns.
- 1
- –
- –
- –
- –
- –
- –
- –
- Research on high-level talent mobility's impact on scientific research performance in the global artificial intelligence field using structural equation model–
A conceptual model of the impact mechanism of mobility on scientific research performance is constructed using the structural equation model and the partial least square algorithm to estimate model parameters.
- –
- –
- Joint Metric Learning on Riemannian Manifold of Global Gaussian Distributions–
Experiments on video based face recognition, object recognition and material classification show that JML is superior to the state-of-the-art metric learning algorithms for Gaussians.
- Early Termination of Intra Mode Decision Based on Most Probable Mode and SATD for H.264/AVC encoding–
Experimental results show that the proposed algorithm effectively reduces the complexity of 4×4 intra prediction while maintaining almost the same coding performance on peak signal-to-noise ratio and bit rate compared with the full search algorithm.
- Cluster Based VDTN Routing Algorithm with Multi-attribute Decision Making–
A cluster based VDTN routing algorithm with multi-attribute decision-making (CR-MA) in urban areas that enables nodes aware of local message distribution and proves its superiority in both message routing effective, performance, and overhead.
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-12. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.