Abstract

Likelihood-based, or explicit, deep generative models use neural networks to construct flexible high-dimensional densities. This formulation directly contradicts the manifold hypothesis, which states that observed data lies on a low-dimensional manifold embedded in high-dimensional ambient space. In this paper we investigate the pathologies of maximum-likelihood training in the presence of this dimensionality mismatch. We formally prove that degenerate optima are achieved wherein the manifold itself is learned but not the distribution on it, a phenomenon we call manifold overfitting. We propose a class of two-step procedures consisting of a dimensionality reduction step followed by maximum-likelihood density estimation, and prove that they recover the data-generating distribution in the nonparametric regime, thus avoiding manifold overfitting. We also show that these procedures enable density estimation on the manifolds learned by implicit models, such as generative adversarial networks, hence addressing a major shortcoming of these models. Several recently proposed methods are instances of our two-step procedures; we thus unify, extend, and theoretically justify a large class of models.

Abstract

Skills or low-level policies in reinforcement learning are temporally extended actions that can speed up learning and enable complex behaviours. Recent work in offline reinforcement learning and imitation learning has proposed several techniques for skill discovery from a set of expert trajectories. While these methods are promising, the number K of skills to discover is always a fixed hyperparameter, which requires either prior knowledge about the environment or an additional parameter search to tune it. We first propose a method for offline learning of options (a particular skill framework) exploiting advances in variational inference and continuous relaxations. We then highlight an unexplored connection between Bayesian nonparametrics and offline skill discovery, and show how to obtain a nonparametric version of our model. This version is tractable thanks to a carefully structured approximate posterior with a dynamically-changing number of options, removing the need to specify K. We also show how our nonparametric extension can be applied in other skill frameworks, and empirically demonstrate that our method can outperform state-of-the-art offline skill learning algorithms across a variety of environments.

Abstract

In text-video retrieval, the objective is to learn a crossmodal similarity function between a text and a video that ranks relevant text-video pairs higher than irrelevant pairs. However, videos inherently express a much wider gamut of information than texts. Instead, texts often capture subregions of entire videos and are most semantically similar to certain frames within videos. Therefore, for a given text, a retrieval model should focus on the text’s most semantically similar video sub-regions to make a more relevant comparison. Yet, most existing works aggregate entire videos without directly considering text. Common text-agnostic aggregations schemes include mean-pooling or self-attention over the frames, but these are likely to encode misleading visual information not described in the given text. To address this, we propose a cross-modal attention model called XPool that reasons between a text and the frames of a video. Our core mechanism is a scaled dot product attention for a text to attend to its most semantically similar frames. We then generate an aggregated video representation conditioned on the text’s attention weights over the frames. We evaluate our method on three benchmark datasets of MSRVTT, MSVD and LSMDC, achieving new state-of-the-art results by up to 12% in relative improvement in Recall@1. Our findings thereby highlight the importance of joint textvideo reasoning to extract important visual cues according to text. Full code and demo can be found at: layer6ailabs.github.io/xpool/.

Abstract

Abstract

Twitter is one of the main information sharing platforms in the world with millions of tweets created daily. To ensure that users get relevant content in their feeds Twitter extensively leverages machine learning-based recommender systems. However, given the large volume of data, all production systems must be both memory and CPU efficient. In the 2021 ACM RecSys challenge Twitter simulates the production environment with a large dataset of almost 1 bilion user-tweet engagements that span a 4 week period. The goal is to accurately predict engagement type, and all models are subject to strict run-time constraints during inference. In this paper we present our approach to the 2021 ACM Recsys challenge. We use a hybrid pipeline and leverage gradient boosting, neural network classifiers and multi-lingual language models to maximize performance. Our approach achieves strong results placing 3’rd on the public leaderboard. We further explore the complexity of language model inference, and show that through distillation it can be possible to run such models in highly constrained production environments.