Yuwei Sun

Researcher at Shanghai Academy of AI for Science
Multimodal Reasoning | NeuroAI

[Yuwei Sun @yweisun]

You can find the paper:
https://t.co/vPt0WDk7fF

The slides and the presentation:
https://t.co/urrLu5dlXs

[Yuwei Sun @yweisun]

Our 2025 CVPR work of Associative Transformer is one of the studies on the global workspace in Transformer architectures, but from a computer vision perspective. We also relate to the Hopfield network to explain how long-term information is maintained in the shared space.

[Anthropic @AnthropicAI]

New Anthropic research: A global workspace in language models.

Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with.

We found a strikingly similar divide inside Claude.

[Dmitry Krotov @DimaKrotov]

This past week several people asked me about how Loop Transformers are related to Energy Transformers.

Energy Transformers are:
👉 looped transformers
👉 energy-based models (EBMs)
👉 Dense Associative Memories — generalized Hopfield nets with superior scaling laws for information storage

That combo is powerful:
👉 looping = iterative refinement = reasoning in the latent space (not in token space)
👉 energy-based = stability of token dynamics
👉 Associative Memory = strong retrieval capabilities

Put together: models that settle into good solutions, not just predict next tokens.

I’ve been especially excited about this class of models for a while. They feel like a promising direction for more stable, interpretable, and memory-rich AI systems.

This week at #ICLR2026 we are presenting NRGPT https://t.co/UkXt54Dzhj, which is a:
👉 a looped transformer
👉 a stable Dense Associative Memory
👉 works great on ListOps and real text

Original Energy Transformer paper: https://t.co/UOfrwZXtOf from NeurIPS 2023

[Ryota Kanai @kanair]

Just posted a new preprint: The Stream of Computation: Temporal Continuity as a Missing Ingredient for Artificial Consciousness. w/ @yweisun & @manuelbaltieri Not a technical paper, but more an essay on the idea of creating AI systems that are always on.
https://t.co/5OlHK6yQsR

[Yuwei Sun @yweisun]

Our paper "MCM: Multi-layer Concept Map for Efficient Concept Learning from Masked Images," has been accepted at #ICLR2025 Workshop on Deep Generative Models! 👏

MCM is an efficient concept learning method that works even with highly masked images.
https://t.co/hDJcz9cqJX

[Yuwei Sun @yweisun]

Our recent paper "Associative Transformer" has been accepted at the main conference of #CVPR2025! 🎉
https://t.co/Ur40LoZc6I
Using sparse bottleneck attention and Hopfield associative memory (hence the name).

Thanks to all my collaborators.
#CVPR2025 #Transformers #hopfield

[Yuwei Sun @yweisun]

Can Models of Human Consciousness Enhance AI Capabilities? https://t.co/Arc7d6W7Rz

[Ryota Kanai @kanair]

Tokyo AI Safety Conference 2025 is happening in April. Let's build a safer AI future together. Some talks from the previous event can be found here:
https://t.co/2HIM21lP4Q

https://t.co/WqeGopFIjy

[Yuwei Sun @yweisun]

Foundations of ML: are deep sparse networks the ultimate answer? https://t.co/cQykVIw3vP via @YouTube

[Ryota Q. Kanai 🌙 @kanair_jp]

意識研究の国際学会ASSCまであと２週間。日本で開催するのは珍しいので（最後は2011年の京都）、これを機会に意識に興味があった人や、現代の意識研究を垣間見たい人は、ぜひ参加してみてください。自分にとっては、一番関心のある研究分野のど真ん中の学会です。
https://t.co/SKo7VcgA0T

[Ryota Kanai @kanair]

Why is counting still so difficult for ChatGPT 4o?

[Yuwei Sun @yweisun]

Meet Disney's robot at #CVPR2024

[Lex Fridman @lexfridman]

Here's my conversation with @CharanRanganath, psychologist & neuroscientist at UC Davis, specializing in human memory. We talk about memory, imagination, deja vu, false memories, why we remember, why we forget, and a lot of other fascinating mysteries about the human mind.

It's here on X in full, and is up on YouTube, Spotify, and everywhere else. Links in comment.

Timestamps:
0:00 - Introduction
1:03 - Experiencing self vs remembering self
14:44 - Creating memories
24:16 - Why we forget
31:53 - Training memory
42:22 - Memory hacks
54:10 - Imagination vs memory
1:03:29 - Memory competitions
1:13:18 - Science of memory
1:28:33 - Discoveries
1:39:37 - Deja vu
1:44:54 - False memories
2:04:59 - False confessions
2:08:45 - Heartbreak
2:16:19 - Nature of time
2:24:00 - Brain–computer interface (BCI)
2:38:04 - AI and memory
2:48:18 - ADHD
2:55:15 - Music
3:05:00 - Human mind

[Yuwei Sun @yweisun]

New blog post: Continual Learning, Compositionality, and Modularity - Part III
Human reasoning identifies abstract patterns from few examples and generalizes them to new inputs through compositional and continual knowledge learning.
https://t.co/HE0EfWdGQ0

#AI #LifelongLearning

[Yuwei Sun @yweisun]

SOTA test accuracy with a ViT-Base backbone model while maintaining great memory efficiency.

[Yuwei Sun @yweisun]

To alleviate forgetting in continual learning, our team recently proposed Remembering Transformer🧠, inspired by the brain's Complementary Learning Systems. Initial results demonstrate new SOTA performances in class-incremental learning tasks.

[Ryota Kanai @kanair]

Check out our new preprint on continual learning. Led by @yweisun.

"Remembering Transformer for Continual Learning"
https://t.co/wN2RcyOeQM

[Yuwei Sun @yweisun]

Remembering Transformer utilizes a mixture-of-adapters architecture and a generative model-based novelty detection mechanism to dynamically route samples to relevant adapters.

[Ryota Kanai @kanair]

We have extended the deadline for ASSC.

[Yuwei Sun @yweisun]

Uncertain neurosymbolic AI is the definitive solution. LLMs possess a substantial amount of world knowledge, yet they lack a framework for stable modularization of knowledge in an online learning context, if the brain indeed operates on the premise of Global Workspace.
#AAAI24

[Ryota Kanai @kanair]

Registration is open!
If you are fascinated by the mysteries of consciousness, join us at ASSC27 in Japan. This is a perfect opportunity to meet with consciousness researchers and visit Tokyo.Abstract submission deadline is Feb 23.
https://t.co/ROdoGJzXnN