Skip to search formSkip to main contentSkip to account menu

Learning to Speak and Act in a Fantasy Text Adventure Game

This work introduces a large-scale crowdsourced text adventure game as a research platform for studying grounded dialogue, and describes the results of training state-of-the-art generative and retrieval models in this setting.
View on ACL (opens in a new tab)

Certified Patch Robustness via Smoothed Vision Transformers

This work demonstrates how using vision transformers enables significantly better certified patch robustness that is also more computationally efficient and does not incur a substantial drop in standard accuracy.
View on IEEE (opens in a new tab)

Missingness Bias in Model Debugging

It is shown how transformer-based architectures can enable a more natural implementation of missingness, which side-steps these issues and improves the reliability of model debugging in practice.
View PDF on arXiv (opens in a new tab)

Combining Diverse Feature Priors

It is shown that models trained with diverse sets of feature priors have less overlapping failure modes, and can thus be combined more effectively, and that jointly training such models on additional (unlabeled) data allows them to correct each other’s mistakes, which leads to better generalization and resilience to spurious correlations.
View PDF on arXiv (opens in a new tab)

Dataset Interfaces: Diagnosing Model Failures Using Controllable Counterfactual Generation

This work introduces dataset interfaces: a framework which allows users to scalably synthesize counterfactual examples from a given dataset, and represents each class from the input dataset as a custom token within the text space of a text-to-image diffusion model.
View PDF on arXiv (opens in a new tab)

Distilling Model Failures as Directions in Latent Space

It is demonstrated that this framework allows us to discover and automatically caption challenging subpopulations within the training dataset and can be used to perform synthetic data augmentation that helps remedy the model’s failure modes.
View PDF on arXiv (opens in a new tab)

When does Bias Transfer in Transfer Learning?

This work demonstrates that bias transfer, or the tendency for biases of the source model to persist even after adapting the model to the target class, can occur even when the target dataset is explicitly de- biased.
View PDF on arXiv (opens in a new tab)

A Data-Based Perspective on Transfer Learning

This work takes a closer look at the role of the source dataset’s composition in transfer learning and presents a framework for probing its impact on downstream performance and demonstrates that removing detrimental datapoints identified by this framework improves transfer learning performance from ImageNet on a variety of target tasks.
View PDF on arXiv (opens in a new tab)

MASA: Motif-Aware State Assignment in Noisy Time Series Data

This work develops motif-aware state assignment (MASA), a method to discover common motifs in noisy time series data and leverage those motifs to more robustly assign states to measurements and is scalable to very large datasets.

A Mechanism for Producing Aligned Latent Spaces with Autoencoders

This work proves that linear and nonlinear autoencoders produce aligned latent spaces by stretching along the left singular vectors of the data and provides an initialization scheme to arbitrarily stretch along the top directions using these networks.
View PDF on arXiv (opens in a new tab)
By clicking accept or continuing to use the site, you agree to the terms outlined in our Privacy Policy (opens in a new tab), Terms of Service (opens in a new tab), and Dataset License (opens in a new tab)