Publications

Many-Shot In-Context Learning
Rishabh Agarwal
Avi Singh
Lei M Zhang
Bernd Bohnet
Stephanie C.Y. Chan
Luis Rosias
Biao Zhang
Ankesh Anand
Zaheer Abbas
Azade Nova
John D Co-Reyes
Eric Chu
Feryal Behbahani
Aleksandra Faust
Hugo Larochelle
Large language models (LLMs) excel at few-shot in-context learning (ICL) -- learning from a few examples provided in context at inference, w… (voir plus)ithout any weight updates. Newly expanded context windows allow us to investigate ICL with hundreds or thousands of examples – the many-shot regime. Going from few-shot to many-shot, we observe significant performance gains across a wide variety of generative and discriminative tasks. While promising, many-shot ICL can be bottlenecked by the available amount of human-generated outputs. To mitigate this limitation, we explore two new settings: (1) "Reinforced ICL" that uses model-generated chain-of-thought rationales in place of human rationales, and (2) "Unsupervised ICL" where we remove rationales from the prompt altogether, and prompts the model only with domain-specific inputs. We find that both Reinforced and Unsupervised ICL can be quite effective in the many-shot regime, particularly on complex reasoning tasks. We demonstrate that, unlike few-shot learning, many-shot learning is effective at overriding pretraining biases, can learn high-dimensional functions with numerical inputs, and performs comparably to supervised fine-tuning. Finally, we reveal the limitations of next-token prediction loss as an indicator of downstream ICL performance.
2024-09-24
NeurIPS.cc/2024/Conference (spotlight)
doi.org
openreview.net
Predicting Future Actions of Reinforcement Learning Agents
Stephen Chung
Scott Niekum
David M. Krueger
2024-09-24
NeurIPS.cc/2024/Conference (poster)
openreview.net
openreview.net
Stress-Testing Capability Elicitation With Password-Locked Models
Ryan Greenblatt
Fabien Roger
Dmitrii Krasheninnikov
David M. Krueger
2024-09-24
NeurIPS.cc/2024/Conference (poster)
doi.org
openreview.net
Textoshop: Interactions Inspired by Drawing Software to Facilitate Text Editing
Damien Masson
Young-Ho Kim
Fanny Chevalier
2024-09-24
ArXiv (prépublication)
doi.org
arxiv.org
The Factorization Curse: Which Tokens You Predict Underlie the Reversal Curse and More
Ouail Kitouni
Niklas Nolte
Adina Williams
Michael G. Rabbat
Diane Bouchacourt
Mark Ibrahim
2024-09-24
NeurIPS.cc/2024/Conference (poster)
openreview.net
openreview.net
Frequency-based View Selection in Gaussian Splatting Reconstruction
Monica Li
Pierre-Yves Lajoie
Giovanni Beltrame
Three-dimensional reconstruction is a fundamental problem in robotics perception. We examine the problem of active view selection to perform… (voir plus) 3D Gaussian Splatting reconstructions with as few input images as possible. Although 3D Gaussian Splatting has made significant progress in image rendering and 3D reconstruction, the quality of the reconstruction is strongly impacted by the selection of 2D images and the estimation of camera poses through Structure-from-Motion (SfM) algorithms. Current methods to select views that rely on uncertainties from occlusions, depth ambiguities, or neural network predictions directly are insufficient to handle the issue and struggle to generalize to new scenes. By ranking the potential views in the frequency domain, we are able to effectively estimate the potential information gain of new viewpoints without ground truth data. By overcoming current constraints on model architecture and efficacy, our method achieves state-of-the-art results in view selection, demonstrating its potential for efficient image-based 3D reconstruction.
2024-09-23
ArXiv (prépublication)
doi.org
arxiv.org
Derivation and validation of indices incorporating vasopressor dose and blood pressure values over time
Alain Gervais
François Lamontagne
Jean-Baptiste Michaud
KJ Neill
Adhikari
Jean-Michel Pagé
Marie-Hélène Masse
Michael O. Harhay
Michaël Chassé
Félix Lamontagne
Katia Laforge
Alexandra Fortin
Marc-André Leclair
Simon Lévesque
Marie-Pier Domingue
Neda Momenzadeh
Martin Vallières
Ruxandra Pinto
Maxime Morin-Lavoie
Francis Carter … (voir 2 de plus)
Félix Camirand Lemyre
MD MSc. François Lamontagne
Rationale The blood pressure value below which the benefits of vasopressors clearly outweigh their disadvantages is uncertain. Objectives Th… (voir plus)e main objective of this analysis was to investigate the statistical properties and potential utility of indices estimating the vasopressor dose-rates as a function of blood pressure values over time. Methods In this single-center observational study, we collected blood pressure values from intensive care unit (ICU) monitors and norepinephrine dose-rates from infusion pumps corresponding to a derivation and a validation cohort. Patients included in each cohort were 18 years or older and received norepinephrine in the ICU. We defined and derived indices corresponding to vasopressor therapy above (>65 mmHg) and below (60 mmHg) targets. We report the distribution of both indices over time from both cohorts as well as their associations with hospital mortal
2024-09-22
medRxiv (prépublication)
doi.org
Not Only the Last-Layer Features for Spurious Correlations: All Layer Deep Feature Reweighting
Humza Wajid Hameed
Geraldin Nanfack
Eugene Belilovsky
Spurious correlations are a major source of errors for machine learning models, in particular when aiming for group-level fairness. It has b… (voir plus)een recently shown that a powerful approach to combat spurious correlations is to re-train the last layer on a balanced validation dataset, isolating robust features for the predictor. However, key attributes can sometimes be discarded by neural networks towards the last layer. In this work, we thus consider retraining a classifier on a set of features derived from all layers. We utilize a recently proposed feature selection strategy to select unbiased features from all the layers. We observe this approach gives significant improvements in worst-group accuracy on several standard benchmarks.
2024-09-22
ArXiv (prépublication)
doi.org
arxiv.org
Self Supervised Dictionary Learning Using Kernel Matching
Shubham Choudhary
Paul Masset
Demba Ba
We introduce a self supervised framework for learning representations in the context of dictionary learning. We cast the problem as a kernel… (voir plus) matching task between the input and the representation space, with constraints on the latent kernel. By adjusting these constraints, we demonstrate how the framework can adapt to different learning objectives. We then formulate a novel Alternate Direction Method of Multipli-ers (ADMM) based algorithm to solve the optimization problem and connect the dynamics to classical alternate minimization techniques. This approach offers a unique way of learning representations with kernel constraints, that enable us implicitly learn a generative map for the data from the learned representations which can have broad applications in representation learning tasks both in machine learning and neuro-science.
2024-09-21
International Workshop on Machine Learning for Signal Processing (publié)
doi.org
Beyond Predictive Algorithms in Child Welfare
Erina Seh-Young Moon
Erin Moon
Devansh Saxena
Tegan Maharaj
Shion Guha
2024-09-20
Graphics Interface (publié)
doi.org
openreview.net
Al content detection in the emerging information ecosystem: new obligations for media and tech companies
Alistair Knott
Dino Pedreschi
Toshiya Jitsuzumi
Susan Leavy
David Eyers
Tapabrata Chakraborti
Andrew Trotman
Sundar Sundareswaran
Ricardo Baeza-Yates
Przemyslaw Biecek
Adrian Weller
Paul D. Teal
Subhadip Basu
Mehmet Haklidir
Virginia Morini
Stuart Russell
Yoshua Bengio
2024-09-20
Ethics and Information Technology (publié)
doi.org
ToxiSight: Insights Towards Detected Chat Toxicity
Zachary Yang
Domenico Tullo
Reihaneh Rabbany
We present a comprehensive explainability dashboard designed for in-game chat toxicity. This dashboard integrates various existing explainab… (voir plus)le AI (XAI) techniques, including token importance analysis, model output visualization, and attribution to the training dataset. It also provides insights through the closest positive and negative examples, facilitating a deeper understanding and potential correction of the training data. Additionally, the dashboard includes word sense analysis—particularly useful for new moderators—and offers free-text explanations for both positive and negative predictions. This multi-faceted approach enhances the interpretability and transparency of toxicity detection models.
2024-09-20
EMNLP/2024/Workshop/BlackBoxNLP (accepté)
openreview.net
openreview.net