Feature visualization Olah et al. (2017) is a set of techniques for neural networks aiming to find inputs that maximize the activation of one or more selected neurons from the same network. Usually, feature visualization is used as a method for model interpretability, where one seeks to understand a neural network by analyzing how much each neuron contributes to a neural network by perceiving the images generated by these techniques. The process of obtaining these inputs is, in a sense, an attempt towards reversing a neural network. Since a neural network is composed by functions that map inputs to outputs, the visual representation of a feature is the input we would have given a target activation for a group of posterior selected neurons. The reversibility of neural networks relates to how well one can reverse the map from the activation of target neurons back to the input neurons Gomez et al. (2017). In most cases, neural networks are not reversible, primarily due to three reasons: (1) the presence of non-reversible activation functions (e.g., ReLU Nair and Hinton (2010)), which means that in general, it is impossible to directly recover the input value x given the output value f(x).

In recent years, neural models learned through self-supervised pretraining on large scale multilingual text or speech data have exhibited promising results for underresourced languages, especially when a relatively large amount of data from related language(s) is available. While the technology has a potential for facilitating tasks carried out in language documentation projects, such as speech transcription, pretraining a multilingual model from scratch for every new language would be highly impractical. We investigate the possibility for adapting an existing multilingual wav2vec 2.0 model for a new language, focusing on actual fieldwork data from a critically endangered tongue: Ainu. Specifically, we (i) examine the feasibility of leveraging data from similar languages also in fine-tuning; (ii) verify whether the model's performance can be improved by further pretraining on target language data. Our results show that continued pretraining is the most effective method to adapt a wav2vec 2.0 model for a new language and leads to considerable reduction in error rates. Furthermore, we find that if a model pretrained on a related speech variety or an unrelated language with similar phonological characteristics is available, multilingual fine-tuning using additional data from that language can have positive impact on speech recognition performance when there is very little labeled data in the target language.

Automatic term extraction (ATE) is a Natural Language Processing (NLP) task that eases the effort of manually identifying terms from domain-specific corpora by providing a list of candidate terms. As units of knowledge in a specific field of expertise, extracted terms are not only beneficial for several terminographical tasks, but also support and improve several complex downstream tasks, e.g., information retrieval, machine translation, topic detection, and sentiment analysis. ATE systems, along with annotated datasets, have been studied and developed widely for decades, but recently we observed a surge in novel neural systems for the task at hand. Despite a large amount of new research on ATE, systematic survey studies covering novel neural approaches are lacking. We present a comprehensive survey of deep learning-based approaches to ATE, with a focus on Transformer-based neural models. The study also offers a comparison between these systems and previous ATE approaches, which were based on feature engineering and non-neural supervised learning algorithms.

Acquiring mathematical skills is considered a key challenge for modern Artificial Intelligence systems. Inspired by the way humans discover numerical knowledge, here we introduce a deep reinforcement learning framework that allows to simulate how cognitive agents could gradually learn to solve arithmetic problems by interacting with a virtual abacus. The proposed model successfully learn to perform multi-digit additions and subtractions, achieving an error rate below 1% even when operands are much longer than those observed during training. We also compare the performance of learning agents receiving a different amount of explicit supervision, and we analyze the most common error patterns to better understand the limitations and biases resulting from our design choices.

With 1.3 billion users, Instagram (IG) has also become a business tool. IG influencer marketing, expected to generate $33.25 billion in 2022, encourages companies and influencers to create trending content. Various methods have been proposed for predicting a post's popularity, i.e., how much engagement (e.g., Likes) it will generate. However, these methods are limited: first, they focus on forecasting the likes, ignoring the number of comments, which became crucial in 2021. Secondly, studies often use biased or limited data. Third, researchers focused on Deep Learning models to increase predictive performance, which are difficult to interpret. As a result, end-users can only estimate engagement after a post is created, which is inefficient and expensive. A better approach is to generate a post based on what people and IG like, e.g., by following guidelines. In this work, we uncover part of the underlying mechanisms driving IG engagement. To achieve this goal, we rely on statistical analysis and interpretable models rather than Deep Learning (black-box) approaches. We conduct extensive experiments using a worldwide dataset of 10 million posts created by 34K global influencers in nine different categories. With our simple yet powerful algorithms, we can predict engagement up to 94% of F1-Score, making us comparable and even superior to Deep Learning-based method. Furthermore, we propose a novel unsupervised algorithm for finding highly engaging topics on IG. Thanks to our interpretable approaches, we conclude by outlining guidelines for creating successful posts.

Synthetic data generation is increasingly important due to privacy concerns. While Autoencoder-based approaches have been widely used for this purpose, sampling from their latent spaces can be challenging. Mixture models are currently the most efficient way to sample from these spaces. In this work, we propose a new approach that models the latent space of an Autoencoder as a simplex, allowing for a novel heuristic for determining the number of components in the mixture model. This heuristic is independent of the number of classes and produces comparable results. We also introduce a sampling method based on probability mass functions, taking advantage of the compactness of the latent space. We evaluate our approaches on a synthetic dataset and demonstrate their performance on three benchmark datasets: MNIST, CIFAR-10, and Celeba. Our approach achieves an image generation FID of 4.29, 13.55, and 11.90 on the MNIST, CIFAR-10, and Celeba datasets, respectively. The best AE FID results to date on those datasets are respectively 6.3, 85.3 and 35.6 we hence substantially improve those figures (the lower is the FID the better). However, AEs are not the best performing algorithms on the concerned datasets and all FID records are currently held by GANs. While we do not perform better than GANs on CIFAR and Celeba we do manage to squeeze-out a non-negligible improvement (of 0.21) over the current GAN-held record for the MNIST dataset.

We study the privatization of distributed learning and optimization strategies. We focus on differential privacy schemes and study their effect on performance. We show that the popular additive random perturbation scheme degrades performance because it is not well-tuned to the graph structure. For this reason, we exploit two alternative graph-homomorphic constructions and show that they improve performance while guaranteeing privacy. Moreover, contrary to most earlier studies, the gradient of the risks is not assumed to be bounded (a condition that rarely holds in practice; e.g., quadratic risk). We avoid this condition and still devise a differentially private scheme with high probability. We examine optimization and learning scenarios and illustrate the theoretical findings through simulations.

Convolutional Neural Networks have been used in a variety of image related applications after their rise in popularity due to ImageNet competition. Convolutional Neural Networks have shown remarkable results in applications including face recognition, moving target detection and tracking, classification of food based on the calorie content and many more. Designing of Convolutional Neural Networks requires experts having a cross domain knowledge and it is laborious, which requires a lot of time for testing different values for different hyperparameter along with the consideration of different configurations of existing architectures. Neural Architecture Search is an automated way of generating Neural Network architectures which saves researchers from all the brute-force testing trouble, but with the drawback of consuming a lot of computational resources for a prolonged period. In this paper, we propose an automated Neural Architecture Search framework DQNAS, guided by the principles of Reinforcement Learning along with One-shot Training which aims to generate neural network architectures that show superior performance and have minimum scalability problem.

We study the statistical decision process of detecting the low-rank signal from various signal-plus-noise type data matrices, known as the spiked random matrix models. We first show that the principal component analysis can be improved by entrywise pre-transforming the data matrix if the noise is non-Gaussian, generalizing the known results for the spiked random matrix models with rank-1 signals. As an intermediate step, we find out sharp phase transition thresholds for the extreme eigenvalues of spiked random matrices, which generalize the Baik-Ben Arous-P\'{e}ch\'{e} (BBP) transition. We also prove the central limit theorem for the linear spectral statistics for the spiked random matrices and propose a hypothesis test based on it, which does not depend on the distribution of the signal or the noise. When the noise is non-Gaussian noise, the test can be improved with an entrywise transformation to the data matrix with additive noise. We also introduce an algorithm that estimates the rank of the signal when it is not known a priori.

Byzantine attacks hinder the deployment of federated learning algorithms. Although we know that the benign gradients and Byzantine attacked gradients are distributed differently, to detect the malicious gradients is challenging due to (1) the gradient is high-dimensional and each dimension has its unique distribution and (2) the benign gradients and the attacked gradients are always mixed (two-sample test methods cannot apply directly). To address the above, for the first time, we propose MANDERA which is theoretically guaranteed to efficiently detect all malicious gradients under Byzantine attacks with no prior knowledge or history about the number of attacked nodes. More specifically, we transfer the original updating gradient space into a ranking matrix. By such an operation, the scales of different dimensions of the gradients in the ranking space become identical. The high-dimensional benign gradients and the malicious gradients can be easily separated. The effectiveness of MANDERA is further confirmed by experimentation on four Byzantine attack implementations (Gaussian, Zero Gradient, Sign Flipping, Shifted Mean), comparing with state-of-the-art defenses. The experiments cover both IID and Non-IID datasets.