最近的一些实证研究表明,重要的机器学习任务,例如训练深神网络,表现出低级别的结构,其中损耗函数仅在输入空间的几个方向上差异很大。在本文中,我们利用这种低级结构来降低基于规范梯度的方法(例如梯度下降(GD))的高计算成本。我们提出的\ emph {低率梯度下降}(lrgd)算法找到了$ \ epsilon $ - approximate的固定点$ p $ - 维功能,首先要识别$ r \ r \ leq p $重要的方向,然后估算真实的方向每次迭代的$ p $维梯度仅通过计算$ r $方向来计算定向衍生物。我们确定强烈凸和非convex目标函数的LRGD的“定向甲骨文复杂性”是$ \ Mathcal {o}(r \ log(1/\ epsilon) + rp) + rp)$ and $ \ Mathcal {o}(R /\ epsilon^2 + rp)$。当$ r \ ll p $时,这些复杂性小于$ \ mathcal {o}的已知复杂性(p \ log(1/\ epsilon))$和$ \ mathcal {o}(p/\ epsilon^2) {\ gd}的$分别在强凸和非凸口设置中。因此,LRGD显着降低了基于梯度的方法的计算成本,以实现足够低级别的功能。在分析过程中,我们还正式定义和表征精确且近似级别函数的类别。
translated by 谷歌翻译
我们研究了非I.I.D的普遍一致性。在线学习的过程中的过程。如果存在一项学习者,则据说随机过程允许普遍一致性,以便在该过程中实现任何可测量的响应函数的消失平均损失。当损失函数无界时,Blanchard等人。表明,唯一承认强烈普遍一致性的进程几乎肯定地占用了有限数量的价值。然而,当损失函数有界时,承认强大的普遍一致性的流程类别更加丰富,其特征可以取决于响应设置(Hanneke)​​。在本文中,我们表明,这类过程与响应设置无关,从而关闭打开的问题(Hanneke,打开问题3)。具体来说,我们表明,允许普遍在线学习的过程对于二进制分类是与具有可数类数量的多字符分类的相同。因此,可以减少具有界限损耗的任何输出设置到二进制分类。我们的减少是建设性和实用性的。实际上,我们表明最近的邻居算法由我们的建筑运输。对于承认强大的普遍学习的过程的二进制分类,我们证明了最近的邻居至少至少学习所有的一系列间隔。
translated by 谷歌翻译
Contrastive representation learning has proven to be an effective self-supervised learning method for images and videos. Most successful approaches are based on Noise Contrastive Estimation (NCE) and use different views of an instance as positives that should be contrasted with other instances, called negatives, that are considered as noise. However, several instances in a dataset are drawn from the same distribution and share underlying semantic information. A good data representation should contain relations between the instances, or semantic similarity and dissimilarity, that contrastive learning harms by considering all negatives as noise. To circumvent this issue, we propose a novel formulation of contrastive learning using semantic similarity between instances called Similarity Contrastive Estimation (SCE). Our training objective is a soft contrastive one that brings the positives closer and estimates a continuous distribution to push or pull negative instances based on their learned similarities. We validate empirically our approach on both image and video representation learning. We show that SCE performs competitively with the state of the art on the ImageNet linear evaluation protocol for fewer pretraining epochs and that it generalizes to several downstream image tasks. We also show that SCE reaches state-of-the-art results for pretraining video representation and that the learned representation can generalize to video downstream tasks.
translated by 谷歌翻译
The optimal layout of a complex system such as aerospace vehicles consists in placing a given number of components in a container in order to minimize one or several objectives under some geometrical or functional constraints. This paper presents an extended formulation of this problem as a variable-size design space (VSDS) problem to take into account a large number of architectural choices and components allocation during the design process. As a representative example of such systems, considering the layout of a satellite module, the VSDS aspect translates the fact that the optimizer has to choose between several subdivisions of the components. For instance, one large tank of fuel might be placed as well as two smaller tanks or three even smaller tanks for the same amount of fuel. In order to tackle this NP-hard problem, a genetic algorithm enhanced by an adapted hidden-variables mechanism is proposed. This latter is illustrated on a toy case and an aerospace application case representative to real world complexity to illustrate the performance of the proposed algorithms. The results obtained using the proposed mechanism are reported and analyzed.
translated by 谷歌翻译
Machine Learning for Source Code (ML4Code) is an active research field in which extensive experimentation is needed to discover how to best use source code's richly structured information. With this in mind, we introduce JEMMA, an Extensible Java Dataset for ML4Code Applications, which is a large-scale, diverse, and high-quality dataset targeted at ML4Code. Our goal with JEMMA is to lower the barrier to entry in ML4Code by providing the building blocks to experiment with source code models and tasks. JEMMA comes with a considerable amount of pre-processed information such as metadata, representations (e.g., code tokens, ASTs, graphs), and several properties (e.g., metrics, static analysis results) for 50,000 Java projects from the 50KC dataset, with over 1.2 million classes and over 8 million methods. JEMMA is also extensible allowing users to add new properties and representations to the dataset, and evaluate tasks on them. Thus, JEMMA becomes a workbench that researchers can use to experiment with novel representations and tasks operating on source code. To demonstrate the utility of the dataset, we also report results from two empirical studies on our data, ultimately showing that significant work lies ahead in the design of context-aware source code models that can reason over a broader network of source code entities in a software project, the very task that JEMMA is designed to help with.
translated by 谷歌翻译
Scaling up neural networks has led to remarkable performance across a wide range of tasks. Moreover, performance often follows reliable scaling laws as a function of training set size, model size, and compute, which offers valuable guidance as large-scale experiments are becoming increasingly expensive. However, previous work on scaling laws has primarily used private data \& models or focused on uni-modal language or vision learning. To address these limitations, we investigate scaling laws for contrastive language-image pre-training (CLIP) with the public LAION dataset and the open-source OpenCLIP repository. Our large-scale experiments involve models trained on up to two billion image-text pairs and identify power law scaling for multiple downstream tasks including zero-shot classification, retrieval, linear probing, and end-to-end fine-tuning. We find that the training distribution plays a key role in scaling laws as the OpenAI and OpenCLIP models exhibit different scaling behavior despite identical model architectures and similar training recipes. We open-source our evaluation workflow and all models, including the largest public CLIP models, to ensure reproducibility and make scaling laws research more accessible. Source code and instructions to reproduce this study will be available at https://github.com/LAION-AI/scaling-laws-openclip
translated by 谷歌翻译
In this paper, we investigate the problem of multi-domain translation: given an element $a$ of domain $A$, we would like to generate a corresponding $b$ sample in another domain $B$, and vice versa. Acquiring supervision in multiple domains can be a tedious task, also we propose to learn this translation from one domain to another when supervision is available as a pair $(a,b)\sim A\times B$ and leveraging possible unpaired data when only $a\sim A$ or only $b\sim B$ is available. We introduce a new unified framework called Latent Space Mapping (\model) that exploits the manifold assumption in order to learn, from each domain, a latent space. Unlike existing approaches, we propose to further regularize each latent space using available domains by learning each dependency between pairs of domains. We evaluate our approach in three tasks performing i) synthetic dataset with image translation, ii) real-world task of semantic segmentation for medical images, and iii) real-world task of facial landmark detection.
translated by 谷歌翻译
The Codex model has demonstrated extraordinary competence in synthesizing code from natural language problem descriptions. However, in order to reveal unknown failure modes and hidden biases, such large-scale models must be systematically subjected to multiple and diverse evaluation studies. In this work, we evaluate the code synthesis capabilities of the Codex model based on a set of 115 Python problem statements from a popular competitive programming portal: HackerRank. Our evaluation shows that Codex is indeed proficient in Python, solving 96% of the problems in a zero-shot setting, and 100% of the problems in a few-shot setting. However, Codex exhibits clear signs of generating memorized code based on our evaluation. This is alarming, especially since the adoption and use of such models could directly impact how code is written and produced in the foreseeable future. With this in mind, we further discuss and highlight some of the prominent risks associated with large-scale models of source code. Finally, we propose a framework for code-synthesis evaluation using variations of problem statements based on mutations.
translated by 谷歌翻译
The spread of misinformation is a prominent problem in today's society, and many researchers in academia and industry are trying to combat it. Due to the vast amount of misinformation that is created every day, it is unrealistic to leave this task to human fact-checkers. Data scientists and researchers have been working on automated misinformation detection for years, and it is still a challenging problem today. The goal of our research is to add a new level to automated misinformation detection; classifying segments of text with persuasive writing techniques in order to produce interpretable reasoning for why an article can be marked as misinformation. To accomplish this, we present a novel annotation scheme containing many common persuasive writing tactics, along with a dataset with human annotations accordingly. For this task, we make use of a RoBERTa model for text classification, due to its high performance in NLP. We develop several language model-based baselines and present the results of our persuasive strategy label predictions as well as the improvements these intermediate labels make in detecting misinformation and producing interpretable results.
translated by 谷歌翻译
Latent variable models such as the Variational Auto-Encoder (VAE) have become a go-to tool for analyzing biological data, especially in the field of single-cell genomics. One remaining challenge is the interpretability of latent variables as biological processes that define a cell's identity. Outside of biological applications, this problem is commonly referred to as learning disentangled representations. Although several disentanglement-promoting variants of the VAE were introduced, and applied to single-cell genomics data, this task has been shown to be infeasible from independent and identically distributed measurements, without additional structure. Instead, recent methods propose to leverage non-stationary data, as well as the sparse mechanism shift assumption in order to learn disentangled representations with a causal semantic. Here, we extend the application of these methodological advances to the analysis of single-cell genomics data with genetic or chemical perturbations. More precisely, we propose a deep generative model of single-cell gene expression data for which each perturbation is treated as a stochastic intervention targeting an unknown, but sparse, subset of latent variables. We benchmark these methods on simulated single-cell data to evaluate their performance at latent units recovery, causal target identification and out-of-domain generalization. Finally, we apply those approaches to two real-world large-scale gene perturbation data sets and find that models that exploit the sparse mechanism shift hypothesis surpass contemporary methods on a transfer learning task. We implement our new model and benchmarks using the scvi-tools library, and release it as open-source software at \url{https://github.com/Genentech/sVAE}.
translated by 谷歌翻译