我们调查与批量无关的归一化产生的性能降解的原因。我们发现层归一化和实例归一化的原型技术均诱导神经网络预激活中的故障模式的外观：（i）层归一化引起往复恒定函数的崩溃;（ii）实例归一化在实例统计中缺乏可变性，表现出富有症状的症状。为了缓解失败模式（i）而不加重失败模式（ii），我们介绍了使用代理分布的激活后标准化的技术“代理归一化”。当与层归一化或组归一化结合时，这种批量独立的归一化会模拟批量标准化的行为，并始终如一地匹配或超过其性能。

Reading comprehension of legal text can be a particularly challenging task due to the length and complexity of legal clauses and a shortage of expert-annotated datasets. To address this challenge, we introduce the Merger Agreement Understanding Dataset (MAUD), an expert-annotated reading comprehension dataset based on the American Bar Association's 2021 Public Target Deal Points Study, with over 39,000 examples and over 47,000 total annotations. Our fine-tuned Transformer baselines show promising results, with models performing well above random on most questions. However, on a large subset of questions, there is still room for significant improvement. As the only expert-annotated merger agreement dataset, MAUD is valuable as a benchmark for both the legal profession and the NLP community.

The increasingly widespread adoption of large language models has highlighted the need for improving their explainability. We present context length probing, a novel explanation technique for causal language models, based on tracking the predictions of a model as a function of the length of available context, and allowing to assign differential importance scores to different contexts. The technique is model-agnostic and does not rely on access to model internals beyond computing token-level probabilities. We apply context length probing to large pre-trained language models and offer some initial analyses and insights, including the potential for studying long-range dependencies. The source code and a demo of the method are available.

Over the past decade, neural networks have been successful at making predictions from biological sequences, especially in the context of regulatory genomics. As in other fields of deep learning, tools have been devised to extract features such as sequence motifs that can explain the predictions made by a trained network. Here we intend to go beyond explainable machine learning and introduce SEISM, a selective inference procedure to test the association between these extracted features and the predicted phenotype. In particular, we discuss how training a one-layer convolutional network is formally equivalent to selecting motifs maximizing some association score. We adapt existing sampling-based selective inference procedures by quantizing this selection over an infinite set to a large but finite grid. Finally, we show that sampling under a specific choice of parameters is sufficient to characterize the composite null hypothesis typically used for selective inference-a result that goes well beyond our particular framework. We illustrate the behavior of our method in terms of calibration, power and speed and discuss its power/speed trade-off with a simpler data-split strategy. SEISM paves the way to an easier analysis of neural networks used in regulatory genomics, and to more powerful methods for genome wide association studies (GWAS).

Recent methods demonstrate that data augmentation using counterfactual knowledge can teach models the causal structure of a task, leading to robust and generalizable models. However, such counterfactual data often has a limited scale and diversity if crowdsourced and is computationally expensive to extend to new perturbation types if generated using supervised methods. To address this, we introduce a new framework called DISCO for automatically generating high-quality counterfactual data at scale. DISCO engineers prompts to generate phrasal perturbations with a large general language model. Then, a task-specific teacher model filters the generation to distill high-quality counterfactual data. We show that learning with this counterfactual data yields a comparatively small student model that is 6% (absolute) more robust and generalizes 5% better across distributions than baselines on various challenging evaluations. This model is also 15% more sensitive in differentiating original and counterfactual examples, on three evaluation sets written by human workers and via human-AI collaboration.

The number of international benchmarking competitions is steadily increasing in various fields of machine learning (ML) research and practice. So far, however, little is known about the common practice as well as bottlenecks faced by the community in tackling the research questions posed. To shed light on the status quo of algorithm development in the specific field of biomedical imaging analysis, we designed an international survey that was issued to all participants of challenges conducted in conjunction with the IEEE ISBI 2021 and MICCAI 2021 conferences (80 competitions in total). The survey covered participants' expertise and working environments, their chosen strategies, as well as algorithm characteristics. A median of 72% challenge participants took part in the survey. According to our results, knowledge exchange was the primary incentive (70%) for participation, while the reception of prize money played only a minor role (16%). While a median of 80 working hours was spent on method development, a large portion of participants stated that they did not have enough time for method development (32%). 25% perceived the infrastructure to be a bottleneck. Overall, 94% of all solutions were deep learning-based. Of these, 84% were based on standard architectures. 43% of the respondents reported that the data samples (e.g., images) were too large to be processed at once. This was most commonly addressed by patch-based training (69%), downsampling (37%), and solving 3D analysis tasks as a series of 2D tasks. K-fold cross-validation on the training set was performed by only 37% of the participants and only 50% of the participants performed ensembling based on multiple identical models (61%) or heterogeneous models (39%). 48% of the respondents applied postprocessing steps.

We study inductive matrix completion (matrix completion with side information) under an i.i.d. subgaussian noise assumption at a low noise regime, with uniform sampling of the entries. We obtain for the first time generalization bounds with the following three properties: (1) they scale like the standard deviation of the noise and in particular approach zero in the exact recovery case; (2) even in the presence of noise, they converge to zero when the sample size approaches infinity; and (3) for a fixed dimension of the side information, they only have a logarithmic dependence on the size of the matrix. Differently from many works in approximate recovery, we present results both for bounded Lipschitz losses and for the absolute loss, with the latter relying on Talagrand-type inequalities. The proofs create a bridge between two approaches to the theoretical analysis of matrix completion, since they consist in a combination of techniques from both the exact recovery literature and the approximate recovery literature.

With an increasing amount of data in the art world, discovering artists and artworks suitable to collectors' tastes becomes a challenge. It is no longer enough to use visual information, as contextual information about the artist has become just as important in contemporary art. In this work, we present a generic Natural Language Processing framework (called ArtLM) to discover the connections among contemporary artists based on their biographies. In this approach, we first continue to pre-train the existing general English language models with a large amount of unlabelled art-related data. We then fine-tune this new pre-trained model with our biography pair dataset manually annotated by a team of professionals in the art industry. With extensive experiments, we demonstrate that our ArtLM achieves 85.6% accuracy and 84.0% F1 score and outperforms other baseline models. We also provide a visualisation and a qualitative analysis of the artist network built from ArtLM's outputs.

Automatic term extraction plays an essential role in domain language understanding and several natural language processing downstream tasks. In this paper, we propose a comparative study on the predictive power of Transformers-based pretrained language models toward term extraction in a multi-language cross-domain setting. Besides evaluating the ability of monolingual models to extract single- and multi-word terms, we also experiment with ensembles of mono- and multilingual models by conducting the intersection or union on the term output sets of different language models. Our experiments have been conducted on the ACTER corpus covering four specialized domains (Corruption, Wind energy, Equitation, and Heart failure) and three languages (English, French, and Dutch), and on the RSDO5 Slovenian corpus covering four additional domains (Biomechanics, Chemistry, Veterinary, and Linguistics). The results show that the strategy of employing monolingual models outperforms the state-of-the-art approaches from the related work leveraging multilingual models, regarding all the languages except Dutch and French if the term extraction task excludes the extraction of named entity terms. Furthermore, by combining the outputs of the two best performing models, we achieve significant improvements.

Although recent deep learning-based calibration methods can predict extrinsic and intrinsic camera parameters from a single image, their generalization remains limited by the number and distribution of training data samples. The huge computational and space requirement prevents convolutional neural networks (CNNs) from being implemented in resource-constrained environments. This challenge motivated us to learn a CNN gradually, by training new data while maintaining performance on previously learned data. Our approach builds upon a CNN architecture to automatically estimate camera parameters (focal length, pitch, and roll) using different incremental learning strategies to preserve knowledge when updating the network for new data distributions. Precisely, we adapt four common incremental learning, namely: LwF , iCaRL, LU CIR, and BiC by modifying their loss functions to our regression problem. We evaluate on two datasets containing 299008 indoor and outdoor images. Experiment results were significant and indicated which method was better for the camera calibration estimation.