Existing methods for distillation use the conventional training approach where all samples participate equally in the process and are thus highly inefficient in terms of data utilization. In this paper, a novel data-efficient approach to transfer the knowledge from a teacher model to a student model is presented. Here, the teacher model uses self-regulation to select appropriate samples for training and identifies their significance in the process. During distillation, the significance information can be used along with the soft-targets to supervise the students. Depending on the use of self-regulation and sample significance information in supervising the knowledge transfer process, three types of distillations are proposed - significance-based, regulated, and hybrid, respectively. Experiments on benchmark datasets show that the proposed methods achieve similar performance as other state-of-the-art methods for knowledge distillation while utilizing a significantly less number of samples.

Meta learning with multiple objectives can be formulated as a Multi-Objective Bi-Level optimization Problem (MOBLP) where the upper-level subproblem is to solve several possible conflicting targets for the meta learner. However, existing studies either apply an inefficient evolutionary algorithm or linearly combine multiple objectives as a single-objective problem with the need to tune combination weights. In this paper, we propose a unified gradient-based Multi-Objective Meta Learning (MOML) framework and devise the first gradient-based optimization algorithm to solve the MOBLP by alternatively solving the lower-level and upper-level subproblems via the gradient descent method and the gradient-based multi-objective optimization method, respectively. Theoretically, we prove the convergence properties of the proposed gradient-based optimization algorithm. Empirically, we show the effectiveness of the proposed MOML framework in several meta learning problems, including few-shot learning, neural architecture search, domain adaptation, and multi-task learning.

For this week's ML practitioner's series, Analytics India Magazine(AIM) got in touch with Théo Viel, a Kaggle GM and a deep learning researcher at a French Startup called Damae Medical. In this interview, Théo shares his rich experiences from an impressive career in the world of algorithms. Théo: I recently graduated from a French Engineering school at which I studied applied Mathematics and Computer Science. I did the Ecole des Ponts ParisTech which is a Grande Ecole: it is a great school to do if you aspire to be a Data Scientist in France. I was introduced to Machine Learning at school, but mostly got into it during my internships, and doing Kaggle competitions.

Modeling geophysical systems as dynamical systems and regressing their vector field from data is a simple way to learn emulators for such systems. We show that when the kernel of these emulators is also learned from data (using kernel flows, a variant of cross-validation), then the resulting data-driven models are not only faster than equation-based models but are easier to train than neural networks such as the long short-term memory neural network. In addition, they are also more accurate and predictive than the latter. When trained on observational data for the global sea-surface temperature, considerable gains are observed by the proposed technique in comparison to classical partial differential equation-based models in terms of forecast computational cost and accuracy. When trained on publicly available re-analysis data for temperatures in the North-American continent, we see significant improvements over climatology and persistence based forecast techniques.

Motivated by the increasing adoption of models which facilitate greater automation in risk management and decision-making, this paper presents a novel Importance Sampling (IS) scheme for measuring distribution tails of objectives modelled with enabling tools such as feature-based decision rules, mixed integer linear programs, deep neural networks, etc. Conventional efficient IS approaches suffer from feasibility and scalability concerns due to the need to intricately tailor the sampler to the underlying probability distribution and the objective. This challenge is overcome in the proposed black-box scheme by automating the selection of an effective IS distribution with a transformation that implicitly learns and replicates the concentration properties observed in less rare samples. This novel approach is guided by a large deviations principle that brings out the phenomenon of self-similarity of optimal IS distributions. The proposed sampler is the first to attain asymptotically optimal variance reduction across a spectrum of multivariate distributions despite being oblivious to the underlying structure. The large deviations principle additionally results in new distribution tail asymptotics capable of yielding operational insights. The applicability is illustrated by considering product distribution networks and portfolio credit risk models informed by neural networks as examples.

Weakly supervised learning has drawn considerable attention recently to reduce the expensive time and labor consumption of labeling massive data. In this paper, we investigate a novel weakly supervised learning problem of learning from similarity-confidence (Sconf) data, where we aim to learn an effective binary classifier from only unlabeled data pairs equipped with confidence that illustrates their degree of similarity (two examples are similar if they belong to the same class). To solve this problem, we propose an unbiased estimator of the classification risk that can be calculated from only Sconf data and show that the estimation error bound achieves the optimal convergence rate. To alleviate potential overfitting when flexible models are used, we further employ a risk correction scheme on the proposed risk estimator. Experimental results demonstrate the effectiveness of the proposed methods.

Generally, self-supervised representation learning trains a feature extractor by solving a pretext task constructed on a large unlabeled dataset. The learned feature extractor yields generic feature representations for other machine learning tasks such as classification. Recent algorithms help a linear classifier to attain classification accuracy comparable to a supervised method from scratch, especially in a few amount of labeled data regime [Newell and Deng, 2020, Hénaff et al., 2020, Chen et al., 2020b]. For example, SwAV [Caron et al., 2020] with ResNet-50 has a top-1 validation accuracy of 75.3% on the ImageNet-1K classification [Deng et al., 2009] compared with 76.5% by using the fully supervised method. InfoNCE [van den Oord et al., 2018] or its modification is a de facto standard objective used in many state-of-the-art self-supervised methods [Logeswaran and Lee, 2018, Bachman et al., 2019, He et al., 2020, Chen et al., 2020a, Hénaff et al., 2020, Caron et al., 2020].

The segmentation of emails into functional zones (also dubbed email zoning) is a relevant preprocessing step for most NLP tasks that deal with emails. However, despite the multilingual character of emails and their applications, previous literature regarding email zoning corpora and systems was developed essentially for English. In this paper, we analyse the existing email zoning corpora and propose a new multilingual benchmark composed of 625 emails in Portuguese, Spanish and French. Moreover, we introduce OKAPI, the first multilingual email segmentation model based on a language agnostic sentence encoder. Besides generalizing well for unseen languages, our model is competitive with current English benchmarks, and reached new state-of-the-art performances for domain adaptation tasks in English.

Open-domain Question Answering models which directly leverage question-answer (QA) pairs, such as closed-book QA (CBQA) models and QA-pair retrievers, show promise in terms of speed and memory compared to conventional models which retrieve and read from text corpora. QA-pair retrievers also offer interpretable answers, a high degree of control, and are trivial to update at test time with new knowledge. However, these models lack the accuracy of retrieve-and-read systems, as substantially less knowledge is covered by the available QA-pairs relative to text corpora like Wikipedia. To facilitate improved QA-pair models, we introduce Probably Asked Questions (PAQ), a very large resource of 65M automatically-generated QA-pairs. We introduce a new QA-pair retriever, RePAQ, to complement PAQ. We find that PAQ preempts and caches test questions, enabling RePAQ to match the accuracy of recent retrieve-and-read models, whilst being significantly faster. Using PAQ, we train CBQA models which outperform comparable baselines by 5%, but trail RePAQ by over 15%, indicating the effectiveness of explicit retrieval. RePAQ can be configured for size (under 500MB) or speed (over 1K questions per second) whilst retaining high accuracy. Lastly, we demonstrate RePAQ's strength at selective QA, abstaining from answering when it is likely to be incorrect. This enables RePAQ to ``back-off" to a more expensive state-of-the-art model, leading to a combined system which is both more accurate and 2x faster than the state-of-the-art model alone.

During the last decade, several research works have focused on providing novel deep learning methods in many application fields. However, few of them have investigated the weight initialization process for deep learning, although its importance is revealed in improving deep learning performance. This can be justified by the technical difficulties in proposing new techniques for this promising research field. In this paper, a survey related to weight initialization techniques for deep algorithms in remote sensing is conducted. This survey will help practitioners to drive further research in this promising field. To the best of our knowledge, this paper constitutes the first survey focusing on weight initialization for deep learning models.