
The Kempner Institute for the Study of Natural and Artificial Intelligence at @Harvard University. RTs ≠ Endorsements
Cambridge, Mass
Joined November 2022
- Tweets1K
- Following360
- Followers4K
- Likes371
What do diffusion models, representation learning, reinforcement learning, and test-time reasoning have in common? Each can be seen as learning an energy function or using one to guide sampling and search.
Excited to share our tutorial on energy-based models with @daibond_alpha!
1/5 EBMs are usually taught as one model family among many. In our new tutorial with @du_yilun, we use them as a single lens on machine learning.
"An Energy-Based Perspective on Machine Learning" arxiv.org/abs/2611.14383
Kempner Institute at Harvard University retweeted
Video models can generate scenes. Geometry models can reconstruct them.
What if one model did both?
UGG generates video, depth and camera motion together, as one scene. (NeurIPS 2026)
Co-first authors: @ZhiyiLi02 and @ruojin8
🌐 ugg-model.github.io
Kempner Institute at Harvard University retweeted
We’ve seen a whole wave of [X]Mimic papers lately, and I might be biased, but this one is easily my favorite 🧡
Introducing ImpedanceMimic: Learned Impedance as Action Space for Force-Accurate Motion Imitation w/ @matthiasjammot
Fixed PD gains can track a motion yet miss its contact forces. ImpedanceMimic learns per-joint stiffness and damping as 129 people walk, run, jump... 🚶🏻♀️🏃🏻♀️
TLDR:
🏆 Beats other action spaces (fixed-PD, torque, muscle) and SOTA (PHC, SimPoE) on motion and force accuracy together
🤖 Transfers without retraining to 22 unseen people and a simulated Unitree G1
🔍 Interpretable gains: stiffness scales with load, damping anticipates unloading
📝 Paper: arxiv.org/abs/2611.19153Proj…
🌐 Project page: impedancemimic.github.io/Imp…
With gracious support from the Kempner Institute at Harvard University. @KempnerInst @hseas
Excited to share UGG, which unifies video generation and 3D reconstruction in a single generative model!
By jointly modeling video, depth and camera pose, we can infer missing parts of a scene, generate new scenes, and improve geometric predictions with inference-time search.
Video models can generate scenes. Geometry models can reconstruct them.
What if one model did both?
UGG generates video, depth and camera motion together, as one scene. (NeurIPS 2026)
Co-first authors: @ZhiyiLi02 and @ruojin8
🌐 ugg-model.github.io
Kempner Institute at Harvard University retweeted
That's a wrap on Memory for Robot Foundation Models at #CoRL2026! Thank you all for the huge interest 🙏
Here is a recap ⬇️
We had 4 invited talks from @KaterinaFragiad , @b_k_bucher, @KarlPertsch and @liu_yuejiang , 6 lightning talks, and 35 accepted papers.
Kempner Institute at Harvard University retweeted
Happy to join Harvard University’s @KempnerInst as a Research Fellow!
I’ll continue to explore energy-based models for generation and reasoning. Excited about the future of the field!
If you work on similar questions, let’s connect here or at NeurIPS in Sydney (presenting GEM).
Excited to share our work on Energy-Based Policy Orchestration that we will be presenting at CoRL!
We present a simple framework to orchestrate different humanoid skills together through an energy value.
yewr.github.io/epo/
Planners operate on task timescales. Feasibility changes on control timescales.
Bridging that gap takes a high-frequency mid-level layer. Introducing EPO: Energy-Based Policy Orchestration #CoRL2026
EPO gives each low-level policy an energy: how feasible a goal is, and how well the policy will reach it. Energies are differentiable and comparable across policies, so EPO optimizes the policy–goal pair at control rate.
Unlike a learned MoE router, EPO is compositional: adding a skill means training one new energy, nothing else. It's also more precise and switches policies 7× less often, pushing the precision–robustness frontier.
📄 Paper: arxiv.org/abs/2611.02546
🌐 Project: yewr.github.io/epo
💻 Code: github.com/YeWR/EPO
@wangyenjen @elvis_hsieh77 @frankzydou @ShiboZhaoSLAM @pabbeel @du_yilun @MIT_LISLab
Kempner Institute at Harvard University retweeted
How do convolutional neural networks benefit from locality and weight sharing? We disentangle their effects, showing that weight sharing can substantially accelerate learning and that matching the receptive field to the target reduces the amount of training data needed. More broadly, we characterize how symmetry and locality shape learning dynamics, and how correlations in the data can change their benefits.
Led by Francesco Mori. Details here: arxiv.org/pdf/2610.30118
Kempner Institute at Harvard University retweeted
🚨New preprint: Interference Beyond Geometry in Concept Extraction
Interference between learned features is almost always measured as geometric overlap. But geometry only tells you what could interfere. Co-activation tells you what does!
We introduce effective interference: geometry × co-activation statistics. It captures not just how features overlap, but how they are actually used together.
Why it matters for sparse autoencoders (SAEs): the encoder produces the codes, so how you build an SAE changes which interference it learns.
This is a work with Valérie Costa (@_valerie_costa_) done at @UAlberta/@AmiiThinks. Valérie is now at @Harvard/@KempnerInst for her PhD studies. 1/
paper: arxiv.org/abs/2609.35351
Kempner Institute at Harvard University retweeted
This is a joint search with the @KempnerInst . Great opportunities and computational resources
Kempner Institute at Harvard University retweeted
EoM has been accepted to #NeurIPS2026 @NeurIPSConf 2026 🎉🚀
See you in Atlanta! ♥️
Paper: arxiv.org/pdf/2606.02859
Imagine a population of machine agents. Each might be strong on certain tasks but fundamentally limited: partial tools, partial observations, finite context, bounded compute.
How can these agents self-orchestrate and self-evolve into stronger collective intelligence to solve tasks beyond any single agent's capability?
Instead of designing the multi-agent system itself, we propose designing the incentives that govern it.
We put agents in an economy. They compete, trade, get wealthy, go bankrupt, and mutate, forming an alive society where coordination and adaptation automatically emerge in a decentralized manner.
Kempner Institute at Harvard University retweeted
LOTION 🧴 has been accepted to #NeurIPS2026!! 🎉🚀
Kempner Institute at Harvard University retweeted
Associate professor of biomedical informatics Marinka Zitnik and colleagues @harvardmed @KempnerInst have developed an artificial intelligence system that analyzes thousands of genes in a patient’s tumor.
#Harvard
harvardmagazine.com/research…
Excited to see ALIGN finally out! 🎊
It’s been fun to explore how speech neuroprostheses can be more robust by enabling reliable decoding across sessions.
Huge thanks to Victoria for this amazing collaboration, and @blsabatini, @mikio_aoi, and @gmishne for their support! 🙏
🎉 ALIGN is at #UAI2026!
We tackle cross-session neural shift in speech BCIs with adversarial learning + temporal augmentation for more robust decoding.
Co-first with @shunnnli, with mentorship and support from @blsabatini, @mikio_aoi & @gmishne
📄 proceedings.mlr.press/v337/z…
Kempner Institute at Harvard University retweeted
Replying to @HarvardMagazine
@HarvardMagazine September issue harvardmagazine.com/2026/09 on COMPASS, our pan-cancer foundation model that predicts immunotherapy response from patients' tumors and identifies the biology driving that response
nature.com/articles/s41591-0…
Wanxiang Shen, Intae Moon, Thinh H. Nguyen, Michelle M. Li, Yepeng Huang, Nitya Nair, Daniel Marbach
Kempner Institute at Harvard University retweeted
Let's rally the computational and theoretical neuroscientists to apply! We truly work across the intelligence spectrum
The application is now open for our #KempnerInstitute Research #Fellowship! Postdocs studying the foundations of #intelligence or applications of #AI are encouraged to apply. Learn more and apply by Oct. 1: bit.ly/4pRzHNb #LLMs #NeuroAI #ML
@ShamKakade6 @blsabatini
Kempner Institute at Harvard University retweeted
Our approach for untangling brain-wide interactions is out today in @NeuroCellPress!
Massive congrats to @mattperich for leading this incredible effort since 2020, and a huge thank you to all of our collaborators.
doi.org/10.1016/j.neuron.202…
Excited to share our new @biorxivpreprint which introduces 𝘾𝙪𝙧𝙧𝙚𝙣𝙩-𝘽𝙖𝙨𝙚𝙙 𝘿𝙚𝙘𝙤𝙢𝙥𝙤𝙨𝙞𝙩𝙞𝙤𝙣 (𝘾𝙐𝙍𝘽𝘿)— an approach for inferring brain-wide interactions using data-constrained recurrent neural network (RNN) models. A thread 👇🏽 (1/8) biorxiv.org/content/10.1101/…
Kempner Institute at Harvard University retweeted
Kempner is hiring fellows! I am a bit biased, but this is an amazing opportunity to hang out with people exploring many interesting questions!
The application is now open for our #KempnerInstitute Research #Fellowship! Postdocs studying the foundations of #intelligence or applications of #AI are encouraged to apply. Learn more and apply by Oct. 1: bit.ly/4pRzHNb #LLMs #NeuroAI #ML
@ShamKakade6 @blsabatini

































