Self-Improvement:
Enabling AI agents to learn from interactions and continually self-improve — actively adapting to new information and novel tasks.
|
Ph.D. Student, Computer Science, Stanford University Email: shgwu [AT] cs [DOT] stanford [DOT] edu 
|
|
|
Hi! My name is Shengguang (I also go by Daniel). I'm a first-year PhD student in Computer Science at Stanford University, currently rotating with Diyi Yang at Stanford Social and Language Technologies (SALT) Lab.
In the last quarter, I worked with Nick Haber at Stanford Autonomous Agents Lab.
|
|
I am interested in developing human-like learning and reasoning skills in machines across domains and modalities. Currently, my work involves the following areas: 
Multimodal Grounding & Reasoning:
Harnessing textual feedback to guide fine-grainded visual perception, and drawing from visual insights to optimize language-based reasoning. |
|
|
(see also Google Scholar)
|
|
TBA, 2025
TL;DR: 3D-Generalist is a generative graphics framework to create 3D environments. Key modules include: 1. diffusion model to generate panoramic images that render the structures of 3D environments; 2. VLA trained via self-improving loop for code generation that refines the environments; 3. another VLA for placing diverse unlabled 3D assets. 3D-Generalist provides a controllabe pipeline to scale up synthetic 3D environment data for embodied AI.
|
|
|
ArXiv, 2025
TL;DR: S-VCO is a novel finetuning method that enhances visual-centric capabilties of VLMs while preserving general performance. Key design is a symmetrical visual contrastive objective that optimizes over visual details while avoiding one-sided "preference" formulation. Across various VLM benchmark domains, S-VCO demonstrates most significant and consistent improvements, with especially strong gains on visually demanding tasks.
|
|
EMNLP-Main, 2024
TL;DR: We proposed novel paradigms for assessing and enhancing social-pragmatic abilities in L(V)LMs. Key results include: 1. open-ended evaluation better reveals LLMs' pragmatic generation as opposed to multiple-choice setup; 2. preferential tuning effectively invokes pragmatic reasoning without compromising generic abilities; 3. improvement of the speaker model's multimodal theory of mind in image referential games.
|
|
ArXiv, 2023
TL;DR: DiverseEvol is an efficient instruction-tuning method that allows the model itself to iteratively sample training subsets to improve its own performance, with a key selection principle of maintaining high diversity in the chosen subsets. Across three datasets and benchmarks, our models, trained on less than 4% of the original dataset, match or improve performance compared with finetuning on full data.
|
|
ArXiv, 2023
TL;DR: We release Qwen, a family of highly-capabale foundation LLMs and Chat-Models. QwenLMs achieve superior performance than baselines (e.g., LLaMA2) of similar sizes on a wide range of benchmarks that measure natural language understanding, reasoning, problem solving, etc. Qwen-72B also outperforms GPT-3.5 on 70% of all tasks.
|
|
ACM-MM, 2024
TL;DR: We present an artifact recovery model that accurately generates images of lost artifacts adhering to historical knowledge. Key designs include: 1. prompt enhancement with archaeological knowledge elicited from LLMs; 2. contrastive learning for textual guidance on correlated historical expertise; 3. visual-semantic constraints on edge and perceptual features for learning intricate visual details.
|
|
EMNLP-Findings, 2023
TL;DR: We introduced a novel non-autoregressive approach to visual storytelling, DiffuVST, which is a diffusion-based LM featuring bidirectional context guidance and multimodal adapters. It directly predicts ground-truth text embeddings from any noisy input, achieving superior performance across NLG metrics at a massively faster inference speed compared to strong autoregressive baselines.
|
Website template from YueYANG1996.github.io. |