Lu Li

The Three Regimes of Offline-to-Online Reinforcement Learning

Offline-to-online reinforcement learning (RL) has emerged as a practical paradigm that leverages offline datasets for pretraining and online… (voir plus) interactions for fine-tuning. However, its empirical behavior is highly inconsistent: design choices of online-fine tuning that work well in one setting can fail completely in another. We propose a stability--plasticity principle that can explain this inconsistency: we should preserve the knowledge of pretrained policy or offline dataset during online fine-tuning, whichever is better, while maintaining sufficient plasticity. This perspective identifies three regimes of online fine-tuning, each requiring distinct stability properties. We validate this framework through a large-scale empirical study, finding that the results strongly align with its predictions in 45 of 63 cases. This work provides a principled framework for guiding design choices in offline-to-online RL based on the relative performance of the offline dataset and the pretrained policy.

2026-06-15

ContinualRL @ Reinforcement Learning Conference (publié)

doi.org

openreview.net

Representation Learning Enables Scalable Multitask Deep Reinforcement Learning

Johan Obando-Ceron

Scaling reinforcement learning (RL) to diverse multitask settings remains a central challenge. While recent advances in model-based RL achie… (voir plus)ve strong performance, they rely on planning and complex training pipelines, making it unclear which components are essential for scalability. We revisit this question and argue that the primary driver of scalable multitask RL is not model-based control, but \emph{representation learning}. In particular, we show that combining predictive, model-based representations with high-capacity value function approximation is sufficient to achieve strong performance, even without planning. We evaluate a simple model-free algorithm, MR.Q, coupled with auxiliary predictive objectives into a scalable actor-critic architecture. This approach outperforms a recent world-model-based method and a range of deep RL baselines across a diverse suite of multitask continuous control tasks, while significantly reducing computational overhead and improving wall-clock efficiency. We observe consistent improvements with increased model capacity and show through ablations that predictive representation learning is critical for performance.

2026-06-03

arXiv (prépublication)

doi.org

arxiv.org

Beyond One-Size-Fits-All: Diagnosis-Driven Online Reinforcement Learning with Offline Priors

Guozheng Ma

Lu Li

Zilin Wang

Pierre-Luc Bacon

Dacheng Tao

Online reinforcement learning (RL) agents increasingly depend on knowledge acquired offline to achieve practical efficiency. Originally stud… (voir plus)ied in offline-to-online RL, this paradigm now spans foundation model post-training and embodied intelligence, with prior types expanding from offline datasets and pre-trained policies to increasingly diverse knowledge sources such as multimodal foundation models and generative world models. Offline priors have become central to how deep RL is developed and deployed. However, this reliance introduces a challenge that the prevailing benchmark-driven paradigm cannot resolve: because prior validity varies across deployments and shifts during training, no single approach to managing it is universally optimal, and benchmark rankings offer limited guidance for real-world deployments. Rather than pursuing universal solutions, we argue that the field should shift to diagnosis-driven tension management, in which deployment-specific evidence guides how the learner relates to its priors throughout training, enabling both flexible and adaptive deployment. We support this position with a framework characterizing how priors reshape online optimization through three functional roles, controlled experiments demonstrating help-or-hurt reversals, cross-domain evidence from foundation model post-training to embodied intelligence, and engagement with five substantive counterarguments.

2026-05-24

DEMO @ International Conference on Machine Learning (poster)

openreview.net

Rotation-Preserving Supervised Fine-Tuning

Mohammad Hamdaqa

Supervised fine-tuning (SFT) improves in-domain performance but can degrade out-of-domain (OOD) generalization. Prior work suggests that thi… (voir plus)s degradation is related to changes in dominant singular subspaces of pretrained weight matrices. However, directly identifying loss-sensitive directions with Hessian or Fisher information is computationally expensive at LLM scale. In this work, we propose preserving projected rotations in pretrained singular subspaces as an efficient proxy for Fisher-sensitive directions, which we call Rotation-Preserving Supervised Fine-Tuning (RPSFT). RPSFT penalizes changes in the projected top-

2026-05-22

CATS @ International Conference on Machine Learning (poster)

doi.org

openreview.net

When 2D Tasks Meet 1D Serialization: On Serialization Friction in Structured Tasks

C.-T.D. Lo

Lu Li

Diji Yang

Tianyu Zhang

Yunkai Zhang

Yoshua Bengio

Yi Zhang

Chung-Hsiang Lo

In the LLM era, many symbolic and structured problems are presented to models through 1D text serialization. Yet some such problems are nati… (voir plus)vely two-dimensional: their relevant relations, such as row--column correspondence or spatial adjacency, are defined by position in a 2D layout rather than by sequential order. This raises a representational question: does preserving the same symbolic entries in a 1D sequence also preserve the relational structure needed for computation? We study this issue through the lens of serialization friction: the representational mismatch in which the same underlying task instances and entries are still present, but relations that depend on layout become implicit under 1D serialization. The study uses a controlled synthetic testbed of three tasks: matrix transpose, Conway's Game of Life, and LU decomposition. In each task, the same instances are presented either as 1D text serialization or as their native 2D layout rendered as an image. Across this testbed, 1D serialization degrades more sharply as task size grows, and errors under serialization exhibit spatially structured patterns, suggesting that this presentation choice is consequential within our testbed. To further interpret these results, we add supplementary analyses that include a within-visual probe and an additional comparison of the two input presentations under the mixed-training transpose setting. These findings suggest that, for layout-defined tasks, reducing inputs to 1D serialization is not a neutral choice of representation.

2026-04-28

arXiv (prépublication)

doi.org

arxiv.org

Stable Gradients for Stable Learning at Scale in Deep Reinforcement Learning

Roger Creus Castanyer

Johan Obando-Ceron

Scaling deep reinforcement learning networks is challenging and often results in degraded performance, yet the root causes of this failure m… (voir plus)ode remain poorly understood. Several recent works have proposed mechanisms to address this, but they are often complex and fail to highlight the causes underlying this difficulty. In this work, we conduct a series of empirical analyses which suggest that the combination of non-stationarity with gradient pathologies, due to suboptimal architectural choices, underlie the challenges of scale. We propose a series of direct interventions that stabilize gradient flow, enabling robust performance across a range of network depths and widths. Our interventions are simple to implement and compatible with well-established algorithms, and result in an effective mechanism that enables strong performance even at large scales. We validate our findings on a variety of agents and suites of environments.

2025-12-02

Conference on Neural Information Processing Systems (Accept (spotlight))

doi.org

openreview.net

Network Sparsity Unlocks the Scaling Potential of Deep Reinforcement Learning

Guozheng Ma

Lu Li

Zilin Wang

Li Shen

Pierre-Luc Bacon

Dacheng Tao

Effectively scaling up deep reinforcement learning models has proven notoriously difficult due to network pathologies during training, motiv… (voir plus)ating various targeted interventions such as periodic reset and architectural advances such as layer normalization. Instead of pursuing more complex modifications, we show that introducing static network sparsity alone can unlock further scaling potential beyond their dense counterparts with state-of-the-art architectures. This is achieved through simple one-shot random pruning, where a predetermined percentage of network weights are randomly removed once before training. Our analysis reveals that, in contrast to naively scaling up dense DRL networks, such sparse networks achieve both higher parameter efficiency for network expressivity and stronger resistance to optimization challenges like plasticity loss and gradient interference. We further extend our evaluation to visual and streaming RL scenarios, demonstrating the consistent benefits of network sparsity.

2025-10-05

Proceedings of the 42nd International Conference on Machine Learning (publié)

doi.org

proceedings.mlr.press

AI for Global Climate Cooperation: Modeling Global Climate Negotiations, Agreements, and Long-Term Cooperation in RICE-N

Tianyu Zhang

Andrew Williams

Phillip Wozny

Kai-Hendrik Cohrs

Koen Ponse

Marco Jiralerspong

Soham Phade

Sunil Srinivasa

Lu Li

Yang Zhang

Prateek Gupta

Erman Acar

Irina Rish

Yoshua Bengio

Stephan Zheng

Comprehensive global cooperation is essential to limit global temperature increases while continuing economic development, e.g., reducing se… (voir plus)vere inequality or achieving long-term economic growth. Achieving long-term cooperation on climate change mitigation with n strategic agents poses a complex game-theoretic problem. For example, agents may negotiate and reach climate agreements, but there is no central authority to enforce adherence to those agreements. Hence, it is critical to design negotiation and agreement frameworks that foster cooperation, allow all agents to meet their individual policy objectives, and incentivize long-term adherence. This is an interdisciplinary challenge that calls for collaboration between researchers in machine learning, economics, climate science, law, policy, ethics, and other fields. In particular, we argue that machine learning is a critical tool to address the complexity of this domain. To facilitate this research, here we introduce RICE-N, a multi-region integrated assessment model that simulates the global climate and economy, and which can be used to design and evaluate the strategic outcomes for different negotiation and agreement frameworks. We also describe how to use multi-agent reinforcement learning to train rational agents using RICE-N. This framework underpinsAI for Global Climate Cooperation, a working group collaboration and competition on climate negotiation and agreement design. Here, we invite the scientific community to design and evaluate their solutions using RICE-N, machine learning, economic intuition, and other domain knowledge. More information can be found on www.ai4climatecoop.org.

2025-04-30

International Conference on Machine Learning (poster)

doi.org

proceedings.mlr.press

MAP: Low-Compute Model Merging with Amortized Pareto Fronts via Quadratic Approximation

Zhiqi Bu

Huan He

Yonghui Wu

Jiang Bian

Yong Chen

Yoshua Bengio

Model merging has emerged as an effective approach to combine multiple single-task models into a multitask model. This process typically inv… (voir plus)olves computing a weighted average of the model parameters without any additional training. Existing model-merging methods focus on enhancing average task accuracy. However, interference and conflicts between the objectives of different tasks can lead to trade-offs during the merging process. In real-world applications, a set of solutions with various trade-offs can be more informative, helping practitioners make decisions based on diverse preferences. In this paper, we introduce a novel and low-compute algorithm, Model Merging with Amortized Pareto Front (MAP). MAP efficiently identifies a Pareto set of scaling coefficients for merging multiple models, reflecting the trade-offs involved. It amortizes the substantial computational cost of evaluations needed to estimate the Pareto front by using quadratic approximation surrogate models derived from a pre-selected set of scaling coefficients. Experimental results on vision and natural language processing tasks demonstrate that MAP can accurately identify the Pareto front, providing practitioners with flexible solutions to balance competing task objectives. We also introduce Bayesian MAP for scenarios with a relatively low number of tasks and Nested MAP for situations with a high number of tasks, further reducing the computational cost of evaluation.

2025-04-22

International Conference on Learning Representations (Accept (Poster))

doi.org

openreview.net

VCR: A Task for Pixel-Level Complex Reasoning in Vision Language Models via Restoring Occluded Text

Ge Zhang

Sai Rajeswar

We introduce Visual Caption Restoration (VCR), a novel vision-language task that challenges models to accurately restore partially obscured … (voir plus)texts using pixel-level hints within images. This task stems from the observation that text embedded in images is intrinsically different from common visual elements and natural language due to the need to align the modalities of vision, text, and text embedded in images. While numerous works have integrated text embedded in images into visual question-answering tasks, approaches to these tasks generally rely on optical character recognition or masked language modeling, thus reducing the task to mainly text-based processing. However, text-based processing becomes ineffective in VCR as accurate text restoration depends on the combined information from provided images, context, and subtle cues from the tiny exposed areas of masked texts. We develop a pipeline to generate synthetic images for the VCR task using image-caption pairs, with adjustable caption visibility to control the task difficulty. With this pipeline, we construct a dataset for VCR called VCR-Wiki using images with captions from Wikipedia, comprising 2.11M English and 346K Chinese entities in both easy and hard split variants. Our results reveal that current vision language models significantly lag behind human performance in the VCR task, and merely fine-tuning the models on our dataset does not lead to notable improvements. We release VCR-Wiki and the data construction code to facilitate future research.

2025-04-22

International Conference on Learning Representations (Accept (Poster))

doi.org

openreview.net

La plateforme Mila Ventures

Boussole des politiques en IA

Séances d'information virtuelles

Publications

La plateforme Mila Ventures

Boussole des politiques en IA

Séances d'information virtuelles

Mots-clés populaires:

Lu Li

Publications