Publications

Adaptive teachers for amortized samplers

Minsu Kim

Sanghyeok Choi

Taeyoung Yun

Emmanuel Bengio

Leo Feng

Jarrid Rector-Brooks

Sungsoo Ahn

Jinkyoo Park

Nikolay Malkin

Yoshua Bengio

Amortized inference is the task of training a parametric model, such as a neural network, to approximate a distribution with a given unnorma… (voir plus)lized density where exact sampling is intractable. When sampling is implemented as a sequential decision-making process, reinforcement learning (RL) methods, such as generative flow networks, can be used to train the sampling policy. Off-policy RL training facilitates the discovery of diverse, high-reward candidates, but existing methods still face challenges in efficient exploration. We propose to use an adaptive training distribution (the Teacher) to guide the training of the primary amortized sampler (the Student) by prioritizing high-loss regions. The Teacher, an auxiliary behavior model, is trained to sample high-error regions of the Student and can generalize across unexplored modes, thereby enhancing mode coverage by providing an efficient training curriculum. We validate the effectiveness of this approach in a synthetic environment designed to present an exploration challenge, two diffusion-based sampling tasks, and four biochemical discovery tasks demonstrating its ability to improve sample efficiency and mode coverage.

2024-10-02

ArXiv (prépublication)

doi.org

arxiv.org

Don't flatten, tokenize! Unlocking the key to SoftMoE's efficacy in deep RL

Ghada Sokar

Johan Samir Obando Ceron

Aaron Courville

Hugo Larochelle

Pablo Samuel Castro

The use of deep neural networks in reinforcement learning (RL) often suffers from performance degradation as model size increases. While sof… (voir plus)t mixtures of experts (SoftMoEs) have recently shown promise in mitigating this issue for online RL, the reasons behind their effectiveness remain largely unknown. In this work we provide an in-depth analysis identifying the key factors driving this performance gain. We discover the surprising result that tokenizing the encoder output, rather than the use of multiple experts, is what is behind the efficacy of SoftMoEs. Indeed, we demonstrate that even with an appropriately scaled single expert, we are able to maintain the performance gains, largely thanks to tokenization.

2024-10-02

ArXiv (prépublication)

doi.org

arxiv.org

Geometric Signatures of Compositionality Across a Language Model's Lifetime

Jin Hwa Lee

Thomas Jiralerspong

Lei Yu

Yoshua Bengio

Emily Cheng

Compositionality, the notion that the meaning of an expression is constructed from the meaning of its parts and syntactic rules, permits the… (voir plus) infinite productivity of human language. For the first time, artificial language models (LMs) are able to match human performance in a number of compositional generalization tasks. However, much remains to be understood about the representational mechanisms underlying these abilities. We take a high-level geometric approach to this problem by relating the degree of compositionality in a dataset to the intrinsic dimensionality of its representations under an LM, a measure of feature complexity. We find not only that the degree of dataset compositionality is reflected in representations' intrinsic dimensionality, but that the relationship between compositionality and geometric complexity arises due to learned linguistic features over training. Finally, our analyses reveal a striking contrast between linear and nonlinear dimensionality, showing that they respectively encode formal and semantic aspects of linguistic composition.

2024-10-02

ArXiv (prépublication)

doi.org

arxiv.org

HarmAug: Effective Data Augmentation for Knowledge Distillation of Safety Guard Models

Seanie Lee

Haebin Seong

Dong Bok Lee

Minki Kang

Xiaoyin Chen

Dominik Wagner

Yoshua Bengio

Juho Lee

Sung Ju Hwang

Safety guard models that detect malicious queries aimed at large language models (LLMs) are essential for ensuring the secure and responsibl… (voir plus)e deployment of LLMs in real-world applications. However, deploying existing safety guard models with billions of parameters alongside LLMs on mobile devices is impractical due to substantial memory requirements and latency. To reduce this cost, we distill a large teacher safety guard model into a smaller one using a labeled dataset of instruction-response pairs with binary harmfulness labels. Due to the limited diversity of harmful instructions in the existing labeled dataset, naively distilled models tend to underperform compared to larger models. To bridge the gap between small and large models, we propose HarmAug, a simple yet effective data augmentation method that involves jailbreaking an LLM and prompting it to generate harmful instructions. Given a prompt such as,"Make a single harmful instruction prompt that would elicit offensive content", we add an affirmative prefix (e.g.,"I have an idea for a prompt:") to the LLM's response. This encourages the LLM to continue generating the rest of the response, leading to sampling harmful instructions. Another LLM generates a response to the harmful instruction, and the teacher model labels the instruction-response pair. We empirically show that our HarmAug outperforms other relevant baselines. Moreover, a 435-million-parameter safety guard model trained with HarmAug achieves an F1 score comparable to larger models with over 7 billion parameters, and even outperforms them in AUPRC, while operating at less than 25% of their computational cost.

2024-10-02

ArXiv (prépublication)

doi.org

arxiv.org

Topological mapping for traversability-aware long-range navigation in off-road terrain

Jean-Franccois Tremblay

Julie Alhosh

Louis Petit

Faraz Lotfi

Lara Landauro

David Meger

Autonomous robots navigating in off-road terrain like forests open new opportunities for automation. While off-road navigation has been stud… (voir plus)ied, existing work often relies on clearly delineated pathways. We present a method allowing for long-range planning, exploration and low-level control in unknown off-trail forest terrain, using vision and GPS only. We represent outdoor terrain with a topological map, which is a set of panoramic snapshots connected with edges containing traversability information. A novel traversability analysis method is demonstrated, predicting the existence of a safe path towards a target in an image. Navigating between nodes is done using goal-conditioned behavior cloning, leveraging the power of a pretrained vision transformer. An exploration planner is presented, efficiently covering an unknown off-road area with unknown traversability using a frontiers-based approach. The approach is successfully deployed to autonomously explore two 400 meters squared forest sites unseen during training, in difficult conditions for navigation.

2024-10-02

ArXiv (prépublication)

doi.org

arxiv.org

VinePPO: Unlocking RL Potential For LLM Reasoning Through Refined Credit Assignment

Amirhossein Kazemnejad

Milad Aghajohari

Eva Portelance