Angehefteter Tweet
tom zollo
185 posts
tom zollo
@SquareZollo
girl/boy dad & adventures @zemelgroup. formerly @barstoolsports and music stuff.
Tarrytown, NY Beigetreten Şubat 2011
670 Folgt1.5K Follower
tom zollo retweetet
Our new preprint on parallelizing training of temporally precise spiking neural networks is out!
We show up to 44x speedups over a conventional sequential baseline. 1/N
English
@jason_lee328 @allen_ai @hungchiayu123 @bqwluckyone @OpenDriveLab Paper: arxiv.org/abs/2507.17383
Github: github.com/thomaspzollo/v…
English
New models include MolmoAct from @jason_lee328 @allen_ai, NORA from @hungchiayu123, and UniVLA from @bqwluckyone @OpenDriveLab
Updated results show that the accuracy vs. calibration relationship may be dependent on model architecture and training objective, and that our approaches to prompt ensembling, action scaling, and time-aware monitoring generalize across models.
English
Nice, I've long suspected something like this: catastrophic forgetting might be an exaggeration afterall.
Huihan Liu@huihan_liu
Even more exciting: VLAs retain knowledge that is seemingly forgotten! A drop in prior task performance does not necessarily indicate complete forgetting of task-relevant knowledge. Instead, this knowledge is still retained, as prior task performance can often be rapidly recovered with minimal finetuning.
English
tom zollo retweetet
New work on continual learning and controllable memory from the Zgroup!
Max S. Bennett@maxsbennett
New paper: Tell Me What To Learn: Generalizing Neural Memory to be Controllable in Natural Language. We built a neural memory system that lets you tell an AI — in natural language — what to remember and what to ignore over time. 🧠👇
English
We propose a framework that combines:
- LLM-based expected information gain for scoring candidate questions
- Heterogeneous GNN propagation to aggregate responses and attributes
- Per-round adaptive respondent selection under explicit budgets
By querying a small, informative subset of individuals, the model infers population-level responses through structured similarity.
English
@ding_ruomeng @zhun_deng @zemelgroup We study a new problem setting: Adaptive Group Elicitation. Under real costs and missing data, the system must dynamically decide:
👉 ❓ Which question to ask
👉 👥 Which individuals to query
👉 🌐 How to leverage population structure to infer unobserved responses
English
@behrouz_ali @maxsbennett @zemelgroup @__YuWang__ Thanks Ali, Titans has been a frequent topic of conversation in our group for a while, such a great paper
English
English
Love this direction! @SquareZollo looks again like great work - congrats!
tom zollo@SquareZollo
Super psyched to finally share our new continual learning paper “Tell Me What To Learn: Generalizing Neural Memory to be Controllable in Natural Language”
English
tom zollo retweetet
New paper: Tell Me What To Learn: Generalizing Neural Memory to be Controllable in Natural Language.
We built a neural memory system that lets you tell an AI — in natural language — what to remember and what to ignore over time.
🧠👇
English
@maxsbennett @zemelgroup also shout out to the great researchers whose work we built on here, eg @behrouz_ali @__YuWang__
English
Was a pleasure working on this with @maxsbennett and @zemelgroup, paper link here: arxiv.org/abs/2602.23201
English