I am a senior undergraduate in Computer Science at Shanghai Jiao Tong University (ACM Honors Class, Zhiyuan College). I am currently a research intern at HuLab, University of Wisconsin–Madison, advised by Prof. Junjie Hu. At SJTU, I work in the Advanced Network Laboratory, advised by Prof. Shengzhong Liu and Prof. Fan Wu. Previously, I was a research intern at Audiocc, advised by Prof. Chenda Li and Prof. Yanmin Qian. Friends call me Hugo.
My research focuses on test-time training and model architecture, especially how a model stores and reads memory over a long context. On the storage side, most deep models give each layer its own memory over time, which keeps time and depth apart. My current work unifies the two into a single recurrence: all layers share one fast-weight state in uTTT, and one last-layer KV history in LastKV. On the reading side, TTG reads a fast-weight memory by generation instead of regression. I have also worked on RL post-training for efficient reasoning (SmartThinker, ICML 2026) and sparse Mixture-of-Experts for speech separation (TF-MoE, INTERSPEECH 2026 Oral).
I am always open to collaboration and academic discussion. Feel free to reach me at hugo0713@sjtu.edu.cn or qhu99@wisc.edu, or on WeChat (ID: HUGO--2025).
I am applying to PhD programs for Fall 2027.
Shares test-time-training fast weights across all layers instead of keeping one private state per layer, which unifies time and depth into a single recurrence and improves both long-context retrieval and novel view synthesis at equal state size.
Reads a fast-weight memory by generating from it instead of regressing to one answer, so when a key has several stored values the memory can recall each of them rather than their average; this improves long-context retrieval and novel view synthesis at the same state size.
Lets every layer read the last layer's KV cache from earlier chunks, which unifies time and depth into a single recurrence for attention. With long-term KV storage cut to roughly a single layer's worth, it still matches or beats standard Transformers on language modeling, novel view synthesis and video generation, and leads the compared architectures on long-context retrieval and most state-tracking tasks.
A sparse Mixture-of-Experts framework that routes experts over time frames and mel bands, raising the capacity of speech-separation models with almost no extra inference cost, so it outperforms models that need far more compute.
An RL method for efficient reasoning that lets the model think only as long as each question needs: during training it estimates the best reasoning length per question and pulls over-long answers toward it, cutting reasoning length by up to half while improving accuracy on benchmarks such as AIME25.
Three projects on long-context memory, all under review at ICLR 2027. Co-first author of uTTT, which shares one fast-weight memory across all layers; second author of TTG, a generative read-out for fast-weight memory; co-author of LastKV, which lets all layers share one last-layer KV history and shrinks long-term KV storage to roughly a single layer's worth while improving state tracking.
Reinforcement learning and post-training for efficient LLM reasoning, including chain-of-thought compression and on-policy distillation.
Co-built JEDI, an AI participant that joins Jitsi meetings and decides when to speak; owned its Kubernetes-native data layer.
Making speech-separation models cheaper to run without shrinking their capacity, via sparse MoE routing over temporal frames and mel-frequency bands.
Course notes and research notes. English translations of the Chinese originals.