Institution profile

Canadian Institute for Advanced Research

Academic institutionnorthamerica · ca
Official website
Research library188linked papers
Opportunities0open roles
Selected work

Representative Papers

torchgfn: A PyTorch GFlowNet library

May 24, 2023arXiv.org

Existing GFlowNet research lacks a unified, scalable PyTorch implementation framework, hindering the development of novel training objectives, integration with diverse environments, and reproducible benchmarking. To address this, we introduce the first modular, object-oriented open-source GFlowNet library built on PyTorch. Our method standardizes environment interfaces and sampler abstractions, enables plug-and-play loss functions—including trajectory balance (TB), detailed balance (DB), and unnormalized balance (UB)—and decouples state-space representation, action policies, and flow parameterizations to facilitate customization and composability. The framework successfully reproduces multiple state-of-the-art results across canonical benchmarks, substantially lowering the barrier for algorithm validation and extension. The codebase is publicly released and has been widely adopted by the research community.

6 citationsRead paper

When Single-Agent with Skills Replace Multi-Agent Systems and When They Fail

Jan 08, 2026arXiv.org

This study investigates whether a single agent can internalize multi-agent collaborative behaviors as a skill selection mechanism to replace conventional multi-agent systems and reduce communication overhead. The authors propose a large language model (LLM)-based skill selection framework augmented with a hierarchical routing mechanism and, drawing inspiration from cognitive science, analyze scalability bottlenecks under expanding skill libraries. They reveal for the first time that LLMs exhibit human-like cognitive capacity limits in skill selection: as the skill library grows, accuracy undergoes a phase-transition-like decline primarily due to semantic interference. Experimental results demonstrate that the proposed approach significantly reduces token consumption and latency while preserving reasoning accuracy, and that hierarchical organization effectively mitigates the performance degradation associated with large-scale skill sets.

4 citationsRead paper

Sufficient conditions for offline reactivation in recurrent neural networks

May 22, 2025International Conference on Learning Representations

Whether noise-driven recurrent neural networks (RNNs) can autonomously replay task-evoked neural activity during input-free resting periods remains an open question—particularly whether task-optimized networks inherently possess offline reactivation capability. Method: We formulate the network dynamics via stochastic differential equations, establish Lyapunov stability conditions, and validate our theory numerically on spatial localization and head-direction estimation tasks. Contribution/Results: We derive the first rigorous mathematical sufficient condition for offline reactivation in RNNs. We prove that denoising dynamics—enabling faithful replay—naturally emerge from smooth stimulus encoding and change-driven optimization, without ad hoc mechanisms. Both theoretical analysis and numerical experiments demonstrate that networks satisfying these optimization principles spontaneously recapitulate online activity patterns during rest, achieving reactivation fidelity exceeding 92%. This reveals offline reactivation as an intrinsic, emergent property of optimally trained recurrent systems, bridging online computation and offline memory consolidation.

3 citationsRead paper

Tight Lower Bounds and Improved Convergence in Performative Prediction

Dec 04, 2024arXiv.org

In the performatively predictive setting, model deployment induces distributional shifts, hindering convergence to a performative stable point. To address this, we propose Affine Risk Minimization (ARM), an algorithm that explicitly models the predictive impact on data distributions by affinely combining historical model parameters—leveraging past training snapshots with learned weights. Theoretically, we establish the first unified convergence upper bound for both existing and new algorithms, proving its tightness; reveal that historical information can break the single-iteration lower-bound barrier; and introduce the first lower-bound analysis framework for ARM-style methods. Empirically, ARM achieves faster convergence and higher accuracy at the performative stable point across multiple benchmarks.

3 citationsRead paper

El Agente Gr\'afico: Structured Execution Graphs for Scientific Agents

Feb 19, 2026

Current large language models lack structured, traceable, and auditable execution mechanisms for scientific workflow automation. This work proposes a single-agent framework that embeds large language model decision-making within a type-safe execution environment and introduces a context management mechanism based on typed symbolic identifiers. By integrating an object-graph mapper with a dynamic knowledge graph, the framework enables structured persistence of computational states and contextual information. The architecture supports robust multi-step parallel computations in tasks such as quantum chemistry, conformational ensemble generation, and metal–organic framework design, demonstrating its scalability, consistency, and efficiency in complex scientific automation scenarios.

1 citationsRead paper
Recent publications

Latest Papers