Large pre-trained, zero-shot capable models have shown considerable success both for standard transfer and adaptation tasks, with particular robustness towards distribution shifts. In addition, subsequent fine-tuning can considerably improve performance on a selected downstream task. However, through naive fine-tuning, these zero-shot models lose their generalizability and robustness towards distribution shifts. This is a particular problem for tasks such as Continual Learning (CL), where continuous adaptation has to be performed as new task distributions are introduced sequentially. In this work, we showcase that where fine-tuning falls short to adapt such zero-shot capable models, simple momentum-based weight interpolation can provide consistent improvements for CL tasks in both memory-free and memory-based settings. In particular, we find improvements of over $+4\%$ on standard CL benchmarks, while reducing the error to the upper limit of jointly training on all tasks at once in parts by more than half, allowing the continual learner to inch closer to the joint training limits.

A grand goal in deep learning research is to learn representations capable of generalizing across distribution shifts. Disentanglement is one promising direction aimed at aligning a models representations with the underlying factors generating the data (e.g. color or background). Existing disentanglement methods, however, rely on an often unrealistic assumption: that factors are statistically independent. In reality, factors (like object color and shape) are correlated. To address this limitation, we propose a relaxed disentanglement criterion - the Hausdorff Factorized Support (HFS) criterion - that encourages a factorized support, rather than a factorial distribution, by minimizing a Hausdorff distance. This allows for arbitrary distributions of the factors over their support, including correlations between them. We show that the use of HFS consistently facilitates disentanglement and recovery of ground-truth factors across a variety of correlation settings and benchmarks, even under severe training correlations and correlation shifts, with in parts over +60% in relative improvement over existing disentanglement methods. In addition, we find that leveraging HFS for representation learning can even facilitate transfer to downstream tasks such as classification under distribution shifts. We hope our original approach and positive empirical results inspire further progress on the open problem of robust generalization.

基于代理的深度度量学习（DML）通过将图像嵌入与班级代表接近的图像（通常相对于它们之间的角度）来学习深度表示。但是，这无视嵌入规范，该规范可以带有其他有益的环境，例如类或图像 - 内在不确定性。此外，基于代理的DML努力学习课堂内部结构。为了立即解决这两个问题，我们引入了基于概率的非各向异性概率代理DML。我们将图像模拟为高超球的定向von mises-fisher（VMF）分布，可以反映图像内部不确定性。此外，我们为类代理提供了非异向von mises-fisher（NIVMF）分布，以更好地表示复杂的类别特异性方差。为了衡量这些模型之间的代理到图像距离，我们开发并研究了多个分布到分配和分布指标。每种框架选择都是由一系列消融研究激励的，这些研究展示了我们对基于代理的DML的概率方法的有益特性，例如不确定性意识，在培训期间较好的梯度以及总体改善的概括性能。后者尤其反映在标准DML基准测试中的竞争性能中，我们的方法可以进行比较，这表明现有的基于代理的DML可以从更概率的治疗中受益匪浅。代码可在github.com/explainableml/probabilistic_deep_metric_learning上找到。

深度度量学习（DML）旨在找到适合于零拍摄传输到先验未知测试分布的表示。但是，公共评估协议仅测试单个固定数据拆分，其中列车和测试类被随机分配。更现实的评估应考虑广泛的分布转变，具有潜在的变化和困难。在这项工作中，我们系统地构建了增加难度的培训 - 测试分裂，并呈现OHLML基准，以在DML中的分发外换档下表征概括。 OODML旨在探讨更具挑战性的泛化性能，多样化的火车到测试分配换档。根据我们的新基准，我们对最先进的DML方法进行了彻底的实证分析。我们发现，虽然泛化趋于难以困难地降解，但随着分布偏移的增加，一些方法在保持性能方面更好。最后，我们提出了几次拍摄的DML作为一种有效的方法，以响应于OHML中呈现的未知测试班次而始终如一地改善泛化。此处可用的代码：https：//github.com/compvis/charracterizing_generalization_in_dml。

Despite the success of large language models (LLMs) in various natural language processing (NLP) tasks, the stored knowledge in these models may inevitably be incomplete, out-of-date, or incorrect. This motivates the need to utilize external knowledge to assist LLMs. Unfortunately, current methods for incorporating external knowledge often require additional training or fine-tuning, which can be costly and may not be feasible for LLMs. To address this issue, we propose a novel post-processing approach, rethinking with retrieval (RR), which retrieves relevant external knowledge based on the decomposed reasoning steps obtained from the chain-of-thought (CoT) prompting. This lightweight approach does not require additional training or fine-tuning and is not limited by the input length of LLMs. We evaluate the effectiveness of RR through extensive experiments with GPT-3 on three complex reasoning tasks: commonsense reasoning, temporal reasoning, and tabular reasoning. Our results show that RR can produce more faithful explanations and improve the performance of LLMs.

Active learning as a paradigm in deep learning is especially important in applications involving intricate perception tasks such as object detection where labels are difficult and expensive to acquire. Development of active learning methods in such fields is highly computationally expensive and time consuming which obstructs the progression of research and leads to a lack of comparability between methods. In this work, we propose and investigate a sandbox setup for rapid development and transparent evaluation of active learning in deep object detection. Our experiments with commonly used configurations of datasets and detection architectures found in the literature show that results obtained in our sandbox environment are representative of results on standard configurations. The total compute time to obtain results and assess the learning behavior can thereby be reduced by factors of up to 14 when comparing with Pascal VOC and up to 32 when comparing with BDD100k. This allows for testing and evaluating data acquisition and labeling strategies in under half a day and contributes to the transparency and development speed in the field of active learning for object detection.

The widely studied task of Natural Language Inference (NLI) requires a system to recognize whether one piece of text is textually entailed by another, i.e. whether the entirety of its meaning can be inferred from the other. In current NLI datasets and models, textual entailment relations are typically defined on the sentence- or paragraph-level. However, even a simple sentence often contains multiple propositions, i.e. distinct units of meaning conveyed by the sentence. As these propositions can carry different truth values in the context of a given premise, we argue for the need to recognize the textual entailment relation of each proposition in a sentence individually. We propose PropSegmEnt, a corpus of over 35K propositions annotated by expert human raters. Our dataset structure resembles the tasks of (1) segmenting sentences within a document to the set of propositions, and (2) classifying the entailment relation of each proposition with respect to a different yet topically-aligned document, i.e. documents describing the same event or entity. We establish strong baselines for the segmentation and entailment tasks. Through case studies on summary hallucination detection and document-level NLI, we demonstrate that our conceptual framework is potentially useful for understanding and explaining the compositionality of NLI labels.

Temporal reasoning is the task of predicting temporal relations of event pairs with corresponding contexts. While some temporal reasoning models perform reasonably well on in-domain benchmarks, we have little idea of the systems' generalizability due to existing datasets' limitations. In this work, we introduce a novel task named TODAY that bridges this gap with temporal differential analysis, which as the name suggests, evaluates if systems can correctly understand the effect of incremental changes. Specifically, TODAY makes slight context changes for given event pairs, and systems need to tell how this subtle contextual change will affect temporal relation distributions. To facilitate learning, TODAY also annotates human explanations. We show that existing models, including GPT-3, drop to random guessing on TODAY, suggesting that they heavily rely on spurious information rather than proper reasoning for temporal predictions. On the other hand, we show that TODAY's supervision style and explanation annotations can be used in joint learning and encourage models to use more appropriate signals during training and outperform across several benchmarks. TODAY can also be used to train models to solicit incidental supervision from noisy sources such as GPT-3 and moves farther towards generic temporal reasoning systems.

Estimating the 6D pose of objects is one of the major fields in 3D computer vision. Since the promising outcomes from instance-level pose estimation, the research trends are heading towards category-level pose estimation for more practical application scenarios. However, unlike well-established instance-level pose datasets, available category-level datasets lack annotation quality and provided pose quantity. We propose the new category level 6D pose dataset HouseCat6D featuring 1) Multi-modality of Polarimetric RGB+P and Depth, 2) Highly diverse 194 objects of 10 household object categories including 2 photometrically challenging categories, 3) High-quality pose annotation with an error range of only 1.35 mm to 1.74 mm, 4) 41 large scale scenes with extensive viewpoint coverage, 5) Checkerboard-free environment throughout the entire scene. We also provide benchmark results of state-of-the-art category-level pose estimation networks.

Code generation models have achieved impressive performance. However, they tend to be brittle as slight edits to a prompt could lead to very different generations; these robustness properties, critical for user experience when deployed in real-life applications, are not well understood. Most existing works on robustness in text or code tasks have focused on classification, while robustness in generation tasks is an uncharted area and to date there is no comprehensive benchmark for robustness in code generation. In this paper, we propose ReCode, a comprehensive robustness evaluation benchmark for code generation models. We customize over 30 transformations specifically for code on docstrings, function and variable names, code syntax, and code format. They are carefully designed to be natural in real-life coding practice, preserve the original semantic meaning, and thus provide multifaceted assessments of a model's robustness performance. With human annotators, we verified that over 90% of the perturbed prompts do not alter the semantic meaning of the original prompt. In addition, we define robustness metrics for code generation models considering the worst-case behavior under each type of perturbation, taking advantage of the fact that executing the generated code can serve as objective evaluation. We demonstrate ReCode on SOTA models using HumanEval, MBPP, as well as function completion tasks derived from them. Interesting observations include: better robustness for CodeGen over InCoder and GPT-J; models are most sensitive to syntax perturbations; more challenging robustness evaluation on MBPP over HumanEval.