Publications

Learning, Fast and Slow: Towards LLMs That Adapt Continually

Rishabh Tiwari

Kusha Sareen

Lakshya A Agrawal

Joseph E. Gonzalez

Matei Zaharia

Kurt Keutzer

Inderjit S Dhillon

Rishabh Agarwal

Devvrit Khatri

Large language models (LLMs) are trained for downstream tasks by updating their parameters (e.g., via RL). However, updating parameters forc… (see more)es them to absorb task-specific information, which can result in catastrophic forgetting and loss of plasticity. In contrast, in-context learning with fixed LLM parameters can cheaply and rapidly adapt to task-specific requirements (e.g., prompt optimization), but cannot by itself typically match the performance gains available through updating LLM parameters. There is no good reason for restricting learning to being in-context or in-weights. Moreover, humans also likely learn at different time scales (e.g., System 1 vs 2). To this end, we introduce a fast-slow learning framework for LLMs, with model parameters as "slow" weights and optimized context as "fast" weights. These fast "weights" can learn from textual feedback to absorb the task-specific information, while allowing slow weights to stay closer to the base model and persist general reasoning behaviors. Fast-Slow Training (FST) is up to 3x more sample-efficient than only slow learning (RL) across reasoning tasks, while consistently reaching a higher performance asymptote. Moreover, FST-trained models remain closer to the base LLM (up to 70% less KL divergence), resulting in less catastrophic forgetting than RL-training. This reduced drift also preserves plasticity: after training on one task, FST trained models adapt more effectively to a subsequent task than parameter-only trained models. In continual learning scenarios, where task domains change on the fly, FST continues to acquire each new task while parameter-only RL stalls.

2026-08-03

CBW @ Conference on Language Modeling (poster)

Merging Adapted Models via Data-Free Covariance Estimation

Marawan Gamal

Derek Tam

Pascal Junior Tikeng Notsawo

Colin Raffel

Guillaume Rabusseau

Model merging provides a way of cheaply combining individual models to produce a model that inherits each individual's capabilities. While s… (see more)ome merging methods can approach the performance of multitask training, they are often heuristically motivated and lack theoretical justification. A principled alternative is to pose model merging as a layer-wise optimization problem that directly minimizes interference between tasks. However, this formulation requires estimating per-layer covariance matrices from data, which may not be available when performing merging. In contrast, many of the heuristically-motivated methods do not require auxiliary data, making them practically advantageous. In this work, we revisit the interference minimization framework and show that, under certain conditions, covariance matrices can be estimated directly from difference matrices, eliminating the need for data while also reducing computational costs. We validate our approach across vision and language benchmarks on models ranging from

2026-08-03

CBW @ Conference on Language Modeling (poster)

A Survey of Agent Memory in the Second Half: Towards Self-Evolving and Long-Horizon Agents

Wei-Chieh Huang

Weizhi Zhang

Yueqing Liang

Yuanchen Bei

Yankai Chen

Tao Feng

xinyu Pan

Zhen Tan

Yu Wang

Tianxin Wei

Shanglin Wu

Ruiyao Xu

Liangwei Yang

Rui Yang

Wooseong Yang

Chin-Yuan Yeh

Hanrong Zhang

Haozhen Zhang

Siqi Zhu

Henry Peng Zou … (see 40 more)

Wanjia Zhao

Song Wang

Wujiang Xu

Zixuan Ke

Zheng Hui

Dawei Li

Yaozu Wu

Langzhou He

Chen Wang

Xiongxiao Xu

Baixiang Huang

Juntao Tan

Shelby Heinecke

Huan Wang

Caiming Xiong

Ahmed Metwally

Jun Yan

Chen-Yu Lee

Hanqing Zeng

Yinglong Xia

Xiaokai Wei

Ali Payani

Yu Wang

Haitong Ma

Wenya Wang

Chenguang Wang

Yu Zhang

Xin Eric Wang

Yongfeng Zhang

Jiaxuan You

Hanghang Tong

Xiao Luo

Xue Liu

Yizhou Sun

Wei Wang

Julian McAuley

James Zou

Jiawei Han

Philip S. Yu

Kai Shu

Research in artificial intelligence is undergoing a paradigm shift from prioritizing model innovations and benchmark scores towards emphasiz… (see more)ing problem definition and rigorous real-world evaluation. As the field enters the "second half," the central challenge becomes real utility in long-horizon, dynamic, and user-dependent settings such as agentic coding, deep research, and computer use, where LLM-based agents face context explosion beyond fixed context windows and must continuously accumulate, manage, and selectively reuse large volumes of information across extended interactions. Memory, with hundreds of papers released in 2025, therefore emerges as the critical solution to fill the utility gap. Beyond serving as passive storage, memory is increasingly the substrate through which agents self-evolve: short-term memory gates which experiences are perceived, selected, and abstracted during execution, while long-term memory accumulates and consolidates them into reusable knowledge and skills, forming the loop through which agents improve from their own experience and sustain continual learning. In this survey, we provide a unified view of foundation agent memory along three dimensions: memory substrate (internal parametric state and external retrieval-augmented stores), cognitive mechanism (sensory, working, episodic, semantic, and procedural), and memory subject (user-centric personalization and agent-centric experience). We then analyze how memory is operated under single- and multi-agent topologies and highlight learning policies over memory operations, showing how memory management itself is becoming a trainable, self-evolving capability that spans reinforcement-learned context curation, experience consolidation at decision time, and the emerging ecosystem of explicit, portable, and shareable agent skills surfaced through agent harnesses, context engineering, and standardized tool-mediation protocols. Finally, we review evaluation benchmarks and metrics for assessing memory utility, and outline various open challenges and future directions.

2026-08-03

Transactions on Machine Learning Research (accepted)

Analytic Planning under Uncertainty with Moment Closure

Shishir Sharma

Doina Precup

Effective model-based reinforcement learning in stochastic environments requires planning that accounts for predictive uncertainty. Propagat… (see more)ing full state distributions analytically offers a principled way to do this, but has traditionally required restrictive policy or reward structures to remain tractable. Consequently, modern deep reinforcement learning has largely retreated to either stochastic sampling, which introduces significant target variance, or deterministic point estimates that ignore predictive covariance entirely. We investigate whether distribution-aware planning is possible without these constraints. Using a quadratic action-value parameterization, we first reduce the Bellman backup to an expectation over the state-value function alone; the key idea is then a compatibility principle between the predictive transition distribution and the value function class, under which this expectation is analytic in the distribution's moments. We instantiate this principle with a Gaussian transition model paired with a radial-basis value function, yielding a closed-form backup that propagates both predictive mean and covariance. Empirically, our approach reduces target variance and yields well-calibrated predictive uncertainty under stochastic observations in continuous control, providing a principled framework for planning with learned distribution models.

2026-08-02

arXiv (preprint)

Can LLMs Faithfully Enact Superforecaster Personas?

Andrew Robert Williams

Evan Jiang

Martin Weiss

Nasim Rahaman

Hugo Larochelle

Superforecasters are people that make forecasts that are statistically better than the average person; they offer an interesting testbed for… (see more) persona modelling. In this preliminary work, we investigate whether large language models (LLMs) can faithfully model superforecaster-like personas making similar forecasts and generating similar rationales behind the forecasts. When we compare LLM-generated responses to the ground truth responses of superforecasters, we find that few-shot prompting generates numerically closer forecasts than instruction-prompting baselines. We also measure the similarity of generated and real rationales, finding that the rationales generated based on examples are judged more similar to those of superforecasters than instruction-prompting baselines. However, an analysis of the reasoning patterns in the rationales shows significant differences between human and LLM forecasters, pointing to a gap between stylistic imitation and deep reasoning similarity. These preliminary results raise several interesting questions about how getting superforecaster-like behaviour from LLMs actually works, and open new avenues to explore for improving the forecasting behaviour of LLMs.

2026-08-02

Social_Sim @ Conference on Language Modeling (poster)

Reasoning with Preference Constraints: A Benchmark for Language Models in Many-to-One Matching Markets

2026-08-01

Transactions on Machine Learning Research (accepted)

Recursive Vision Language Models for General Symbolic Reasoning

Omid Nejati Manzari

Guillaume Lajoie

Hassan Rivaz

Hard symbolic-reasoning tasks such as Sudoku, maze pathfinding, and ARC remain challenging for LLMs due to their fixed-depth autoregressive … (see more)reasoning, which limits systematic search, refinement, and backtracking. While recursive models such as Hierarchical Reasoning Model (HRM) and Tiny Recursive Model (TRM) address this limitation through iterative latent-state refinement, they are typically task-specific and do not leverage pretrained language priors. We propose R-Qwen, a recursive reasoning framework built upon a pretrained Qwen backbone. R-Qwen repeatedly refines a candidate solution through programmatic self-recursion and deep supervision, combining the structured iterative computation of recursive models with the linguistic and reasoning priors of pretrained LLMs. We further adapt Hierarchical Supervision Weighting (HSW) to autoregressive models by exponentially weighting losses across recursive steps. HSW reduces gradient variance by at least 50\%, improves the signal-to-noise ratio of stochastic gradients, and accelerates convergence. Across eight challenging benchmarks, R-Qwen consistently outperforms prior recursive reasoning models and substantially larger LLMs while using a comparable number of trainable parameters. Notably, on ARC-AGI dataset, our model achieves a 27.6\% improvement over the baseline, highlighting the effectiveness of recursive refinement for general symbolic reasoning. These results suggest that recursive reasoning mechanisms and pretrained language model priors are complementary approaches for improving symbolic puzzle-solving. Code and models will be released after acceptance.

2026-08-01

arXiv (preprint)

Inferring brain-wide interactions using data-constrained recurrent neural network models

Matthew G. Perich

Charlotte Arlt

Sofia Soares

Siyan Zhou

Manuel Beiran

Aaron S. Andalman

Tyler Benster

Megan E. Young

Clayton P. Mosher

Juri Minxha

Eugene Carter

Ueli Rutishauser

Peter H. Rudebeck

Christopher D. Harvey

Karl Deisseroth

Kanaka Rajan

2026-07-31

Neuron (published)

Nonlinear Laplacians Improve Signed-Directed Graph Learning

Ali Parviz

Yuichi Yoshida

While signed-directed graphs have been studied using linear Laplacians in the design of graph neural networks, relatively little research ha… (see more)s focused on developing non-linear Laplacian operators for such networks. We introduce a non-linear Laplacian operator specific to signed and directed networks (NLSD). This non-linear operator extends the concepts of the signed Laplacian for signed graphs and the Laplacian for directed graphs. The NLSD calculates node-specific potentials based on features More precisely, if the potential discrepancy is not aligned with the edge direction, we ignore it (and vice versa) leveraging message-passing techniques only across edges where potential discrepancies align with the edge's direction. Utilizing this novel operator, we propose an efficient spectral GNN framework (NLSD-GNN). We conducted comprehensive evaluations focusing on node classification and link prediction, examining scenarios involving signed, directional, or both types of information. Our findings reveal that this spectral GNN framework not only integrates signed and directional data effectively but also achieves superior performance across diverse datasets.

2026-07-31

arXiv (preprint)

De novo L-(+)-tartaric acid biosynthesis in multi-modular engineered yeasts

Xuan Zhou

Jiaheng Hou

Zikai Wang

Zhendong Li

Yang Li

Xitong Li

Xianhao Xu

Yanfeng Liu

Jianghua Li

Guocheng Du

Dacheng Ma

J. Tang

Jian Chen

Xueqin Lv

Long Liu

L-(+)-tartaric acid (L-TA) is a high-value chiral organic acid essential for food and pharmaceuticals. Despite its industrial importance, su… (see more)stainable green production is constrained by the lack of a fully defined biosynthetic pathway. Here, we report the de novo biosynthesis of L-TA in Saccharomyces cerevisiae through reaction-guided enzyme mining, experimental validation, and Enzyme Commission-specific Catalytic Hybrid Optimizer (ECHO)-assisted enzyme prioritization. We first elucidate the elusive two-step conversion from precursor 5-keto-D-gluconic acid (5-KGA) to L-TA, catalyzed by transketolase (TK) and succinate semialdehyde dehydrogenase (SSDH). To optimize this critical step, we develop the ECHO. This multimodal framework integrates sequence, substrate, and pocket-aware structural information to identify high-performance TK-SSDH pairs. By integrating this pathway with de novo precursor synthesis, cofactor engineering, and semi-rational protein engineering, a final L-TA titer of 6.59 mg L−1 was achieved in a 5-L bioreactor. By connecting computational mining and metabolic assembly through a multi-module engineering strategy, our study establishes a green platform for L-TA production and demonstrates an effective workflow for synthetic pathway design. L-(+)-tartaric acid (L-TA) is a high-value chiral organic acid for food and pharmaceuticals. Here the authors produce L-TA in S. cerevisiae through reaction-guided enzyme mining and Enzyme Commission-specific Catalytic Hybrid Optimizer (ECHO)-assisted enzyme prioritization.

2026-07-31

Nature Communications (published)

Preparation of MgAl2O4-reinforced magnesium composite refractories from high silicon magnesite tailings: α-Al2O3 /AlN reaction mechanism and thermal shock resistance

Sheng Wu

Qingdong Hou

Xudong Luo

Jun Ding

Cairan Wang

Jinfan Xu

2026-07-31

Journal of Alloys and Compounds (published)