A5102007400- Topic Modeling
- Natural Language Processing Techniques
- Software Engineering Research
- Multimodal Machine Learning Applications
- Advanced Malware Detection Techniques
- Advanced Image Processing Techniques
- Advanced Graph Neural Networks
- Speech Recognition and Synthesis
- Explainable Artificial Intelligence (XAI)
- Expert finding and Q&A systems
- Machine Learning and Data Classification
- Pluripotent Stem Cells Research
- Face recognition and analysis
- Semantic Web and Ontologies
- Speech and dialogue systems
- Generative Adversarial Networks and Image Synthesis
- Software Testing and Debugging Techniques
- Advanced Neural Network Applications
- Text Readability and Simplification
- Domain Adaptation and Few-Shot Learning
- Dietary Effects on Health
- Adversarial Robustness in Machine Learning
- Machine Learning and Algorithms
- RNA and protein synthesis mechanisms
- Opportunistic and Delay-Tolerant Networks
National University of Singapore
2023-2024
Beijing Jiaotong University
2017-2023
East China Normal University
2022-2023
Alibaba Group (China)
2023
Institute for Infocomm Research
2023
Agency for Science, Technology and Research
2023
NetEase (China)
2023
Sichuan University
2022
Institute of Genetics and Developmental Biology
2017
State Key Laboratory of Plant Genomics
2017
Large language models (LLMs) have a wealth of knowledge that allows them to excel in various Natural Language Processing (NLP) tasks. Current research focuses on enhancing their performance within existing knowledge. Despite vast knowledge, LLMs are still limited by the amount information they can accommodate and comprehend. Therefore, ability understand own limitations unknows, referred as self-knowledge, is paramount importance. This study aims evaluate LLMs' self-knowledge assessing...
The differentiation of pluripotent stem cells (PSCs) into diverse functional cell types provides a promising solution to support drug discovery, disease modeling, and regenerative medicine. However, is currently limited by the substantial line-to-line batch-to-batch variabilities, which severely impede progress scientific research manufacturing products. For instance, PSC-to-cardiomyocyte (CM) vulnerable inappropriate doses CHIR99021 (CHIR) that are applied in initial stage mesoderm...
Translational efficiency is subject to extensive regulation. However, the factors influencing such regulation are poorly understood. In Caenorhabditis elegans , 62% of genes trans- spliced a specific leader (SL1), which replaces part native 5′ untranslated region (5′ UTR). Given pivotal role UTR plays in translational efficiency, we hypothesized that SL1 splicing functions regulate efficiency. With genome-wide analysis on Ribo-seq data, polysome profiling experiments, and CRISPR-Cas9–based...
Direct conversion of fibroblasts into induced cardiomyocytes (iCMs) holds promising potential to generate functional for drug development and clinical applications, especially direct in situ heart regeneration by delivery reprogramming genes adult cardiac injured hearts. For a decade, many cocktails transcription factors have been developed iCMs from different tissues vitro some were applied vivo . Here, we aimed develop genetic that induce directly cultured isolated mice with myocardial...
Neural Code Intelligence -- leveraging deep learning to understand, generate, and optimize code holds immense potential for transformative impacts on the whole society. Bridging gap between Natural Language Programming Language, this domain has drawn significant attention from researchers in both research communities over past few years. This survey presents a systematic chronological review of advancements intelligence, encompassing 50 representative models their variants, more than 20...
Graphical User Interface (GUI) agents are designed to automate complex tasks on digital devices, such as smartphones and desktops. Most existing GUI interact with the environment through extracted structured data, which can be notably lengthy (e.g., HTML) occasionally inaccessible desktops). To alleviate this issue, we propose a novel visual agent -- SeeClick, only relies screenshots for task automation. In our preliminary study, have discovered key challenge in developing agents: grounding...
Recent advancements in Chain-of-Thought prompting have facilitated significant breakthroughs for Large Language Models (LLMs) complex reasoning tasks. Current research enhances the performance of LLMs by sampling multiple chains and ensembling based on answer frequency. However, this approach fails scenarios where correct answers are minority. We identify as a primary factor constraining capabilities LLMs, limitation that cannot be resolved solely predicted answers. To address shortcoming,...
Recently, Chain-of-Thought (CoT) prompting has delivered success on complex reasoning tasks, which aims at designing a simple prompt like ``Let's think step by step'' or multiple in-context exemplars with well-designed rationales to elicit Large Language Models (LLMs) generate intermediate steps. However, the generated often come mistakes, making unfactual and unfaithful chains. To mitigate this brittleness, we propose novel Chain-of-Knowledge (CoK) prompting, where aim eliciting LLMs...
Long-term live-cell imaging technology has emerged in the study of cell culture and development, it is expected to elucidate differentiation or reprogramming morphology cells dynamic process interaction between cells. There are some advantages this technique: noninvasive, high-throughput, low-cost, can help researchers explore phenomena that otherwise difficult observe. Many challenges arise real-time process, for example, low-quality micrographs often obtained due unavoidable human factors...
Large language models (LLMs) have shown increasing power on various natural processing (NLP) tasks. However, tuning these for downstream tasks usually needs exorbitant costs or is unavailable due to commercial considerations. Recently, black-box has been proposed address this problem by optimizing task-specific prompts without accessing the gradients and hidden representations. most existing works yet fully exploited potential of gradient-free optimization under scenario few-shot learning....
Large language models (LLMs) have a wealth of knowledge that allows them to excel in various Natural Language Processing (NLP) tasks. Current research focuses on enhancing their performance within existing knowledge. Despite vast knowledge, LLMs are still limited by the amount information they can accommodate and comprehend. Therefore, ability understand own limitations unknows, referred as self-knowledge, is paramount importance. This study aims evaluate LLMs' self-knowledge assessing...
Code pre-trained models (CodePTMs) have recently demonstrated a solid capacity to process various software intelligence tasks, e.g., code clone detection, translation, and summarization. The current mainstream method that deploys these downstream tasks is fine-tune them on individual which generally costly needs sufficient data for large models. To tackle the issue, in this paper, we present TransCoder, unified Transferable fine-tuning strategy representation learning. Inspired by human...
Large Language Models (LLMs) are evolving at an unprecedented pace and have exhibited considerable capability in the realm of natural language processing (NLP) with world knowledge. Benefiting from ultra-large-scale training corpora, a single LLM can manage typical NLP tasks competently. However, its performance executing reasoning is still confined by limitations internal representations. To push this boundary further, we introduce Corex paper, suite novel general-purpose strategies that...
In this paper, we introduce HugNLP, a unified and comprehensive library for natural language processing (NLP) with the prevalent backend of Hugging Face Transformers, which is designed NLP researchers to easily utilize off-the-shelf algorithms develop novel methods user-defined models tasks in real-world scenarios. HugNLP consists hierarchical structure including models, processors applications that unifies learning process pre-trained (PLMs) on different tasks. Additionally, present some...
In urban vehicular ad hoc networks (VANETs), the intersection-based routing scheme has represented its greater applicability and better efficiency to adapt high constrained mobility. How make an accurate decision for street selection is a challenging issue due rapid topology changes in VANETs. this paper, we propose microscopic mechanism based on intersection records (MMIR) which vehicle nodes maintain update table with every passing vehicle’s individual information. By analyzing processing...
Code pre-trained models (CodePTMs) have recently demonstrated significant success in code intelligence. To interpret these models, some probing methods been applied. However, fail to consider the inherent characteristics of codes. In this paper, address problem, we propose a novel method CAT-probing quantitatively how CodePTMs attend structure. We first denoise input sequences based on token types pre-defined by compilers filter those tokens whose attention scores are too small. After that,...
Graphical User Interface (GUI) agents powered by Vision-Language Models (VLMs) have demonstrated human-like computer control capability. Despite their utility in advancing digital automation, a critical bottleneck persists: collecting high-quality trajectory data for training. Common practices such rely on human supervision or synthetic generation through executing pre-defined tasks, which are either resource-intensive unable to guarantee quality. Moreover, these methods suffer from limited...
Over the past few years, we have witnessed remarkable advancements in Code Pre-trained Models (CodePTMs). These models achieved excellent representation capabilities by designing structure-based pre-training tasks for code. However, how to enhance absorption of structural knowledge when fine-tuning CodePTMs still remains a significant challenge. To fill this gap, paper, present Structure-aware Fine-tuning (SAT), novel structure-enhanced and plug-and-play method CodePTMs. We first propose...
One of the primary driving forces contributing to superior performance Large Language Models (LLMs) is extensive availability human-annotated natural language data, which used for alignment fine-tuning. This inspired researchers investigate self-training methods mitigate reliance on human annotations. However, current success has been primarily observed in scenarios, rather than increasingly important neural-symbolic scenarios. To this end, we propose an environment-guided framework named...
In recent years, the rapid increase in scientific papers has overwhelmed traditional review mechanisms, resulting varying quality of publications. Although existing methods have explored capabilities Large Language Models (LLMs) for automated reviewing, their generated contents are often generic or partial. To address issues above, we introduce an paper reviewing framework SEA. It comprises three modules: Standardization, Evaluation, and Analysis, which represented by models SEA-S, SEA-E,...