LangGPT

LangGPT is a personified structural prompt framework. It presents a structured, flexible and extensible framework for prompt creation with reference to the design ideas of object-oriented programming languages. It gives identity to large language models, greatly improving their performance on specific tasks.

URL: https://github.com/langgptai/LangGPT
Paper: LangGPT: Rethinking Structured Reusable Prompt Design Framework for LLMs from the Programming Language
Contribution: Core Contributor & Co-builder, responsible for theoretical system refinement and paper writing.

Minstrel

Minstrel is an automatic prompt optimization (or generation) tool. It decomposes the prompt generation task and automates the generation of structural LangGPT prompts through collaboration between generative agents belonging to three working groups.

URL: https://github.com/langgptai/Minstrel
Paper: Minstrel: Structural Prompt Generation with Multi-Agents Coordination for Non-AI Experts
Contribution: Person in Charge, hosting the design, development, and testing of Minstrel.

T-COL

T-COL is a counterfactual explanation generation method. It utilizes a link tree structure to select several local feature subsets through a partitioning strategy, which are eventually stitched together into a counterfactual interpretation.

URL: https://github.com/sci-m-wang/T-COL
Paper: T-COL: Generating Counterfactual Explanations for General User Preferences on Variable Machine Learning Systems
Contribution: Person in Charge, hosting the design, development, and testing of T-COL.

MM-Bigbench

MM-BigBench, with a range of diverse metrics to thoroughly evaluate different models and instructions, including the Best Performance metric, the Mean Relative Gain metric, the Stability metric, and the Adaptability metric.

URL: https://github.com/declare-lab/MM-BigBench
Paper: MM-BigBench: Evaluating Multimodal Models on Multimodal Content Comprehension Tasks
Contribution: Key Participant, evaluating multimodal large language models such as VPGTrans, LlaMa-Adapter, LlaVa, etc.

PICA

Although traditional language models play a good role in giving help, there is a problem of emotional participation deficiency, since long suggestive replies are against the goal of empathy. To solve the problem, we put forward the multi-turn dialogue model PICA, a chatbot that can empathize and rely on the emotion required.

URL: https://github.com/NEU-DataMining/PICA
Contribution: Key Participant, participating in data construction and supervised fine-tuning.

Funds

A Study of Interpretable Dialogue Generation Techniques for Emotion Awareness (No.62172086)
Student Participant; Funding: ￥600,000
National Natural Science Foundation of China (NSFC)

Internet of Things Teaching Experiment Box (No.201810217141)
Person in Charge; Funding: ￥10,000
National Student Innovation Training Projects

Smart Home System Based on Internet of Things (No.201810217140)
Key Participants; Funding: ￥15,000
National Student Innovation Training Projects

Adsorption Type Underwater Hull Surface Inspection Robot (No.201810217024)
Key Participants; Funding: ￥12,000
National Student Innovation Training Projects

Ant Travel (No.201810217256)
Key Participants; Funding: ￥27,000
National Student Entrepreneurship Training Projects