It is challenging to train deep neural networks robustly on the industrial-level data, since labels of such data are heavily noisy, and their label generation processes are normally agnostic. To handle these issues, by using the memorization effects of deep neural networks, we may train deep neural networks on the whole dataset only the first few iterations. Then, we may employ early stopping or the small-loss trick to train them on selected instances. However, in such training procedures, deep neural networks inevitably memorize some noisy labels, which will degrade their generalization. In this paper, we propose a meta algorithm called Pumpout to overcome the problem of memorizing noisy labels. By using scaled stochastic gradient ascent, Pumpout actively squeezes out the negative effects of noisy labels from the training model, instead of passively forgetting these effects. We leverage Pumpout to upgrade two representative methods: MentorNet and Backward Correction. Empirical results on benchmark datasets demonstrate that Pumpout can significantly improve the robustness of representative methods.

Thanks to the availability of large scale digital datasets and massive amounts of computational power, deep learning algorithms can learn representations of data by exploiting multiple levels of abstraction. These machine learning methods have greatly improved the state-of-the-art in many challenging cognitive tasks, such as visual object recognition, speech processing, natural language understanding and automatic translation. In particular, one class of deep learning models, known as deep belief networks, can discover intricate statistical structure in large data sets in a completely unsupervised fashion, by learning a generative model of the data using Hebbian-like learning mechanisms. Although these self-organizing systems can be conveniently formalized within the framework of statistical mechanics, their internal functioning remains opaque, because their emergent dynamics cannot be solved analytically. In this article we propose to study deep belief networks using techniques commonly employed in the study of complex networks, in order to gain some insights into the structural and functional properties of the computational graph resulting from the learning process.

Generative networks have made it possible to generate meaningful signals such as images and texts from simple noise. Recently, generative methods based on GAN and VAE were developed for graphs and graph signals. However, some of these methods are complex as well as difficult to train and fine-tune. This work proposes a graph generation model that uses a recent adaptation of Mallat's scattering transform to graphs. The proposed model is naturally composed of an encoder and a decoder. The encoder is a Gaussianized graph scattering transform. The decoder is a simple fully connected network that is adapted to specific tasks, such as link prediction, signal generation on graphs and full graph and signal generation. The training of our proposed system is efficient since it is only applied to the decoder and the hardware requirement is moderate. Numerical results demonstrate state-of-the-art performance of the proposed system for both link prediction and graph and signal generation. These results are in contrast to experience with Euclidean data, where it is difficult to form a generative scattering network that performs as well as state-of-the-art methods. We believe that this is because of the discrete and simpler nature of graph applications, unlike the more complex and high-frequency nature of Euclidean data, in particular, of some natural images.

We propose a new architecture called Memory-Augmented Encoder-Solver (MAES) that enables transfer learning to solve complex working memory tasks adapted from cognitive psychology. It uses dual recurrent neural network controllers, inside the encoder and solver, respectively, that interface with a shared memory module and is completely differentiable. We study different types of encoders in a systematic manner and demonstrate a unique advantage of multi-task learning in obtaining the best possible encoder. We show by extensive experimentation that the trained MAES models achieve task-size generalization, i.e., they are capable of handling sequential inputs 50 times longer than seen during training, with appropriately large memory modules. We demonstrate that the performance achieved by MAES far outperforms existing and well-known models such as the LSTM, NTM and DNC on the entire suite of tasks.

We propose a learning-based filter that allows us to directly modify a synthetic speech waveform into a natural speech waveform. Speech-processing systems using a vocoder framework such as statistical parametric speech synthesis and voice conversion are convenient especially for a limited number of data because it is possible to represent and process interpretable acoustic features over a compact space, such as the fundamental frequency (F0) and mel-cepstrum. However, a well-known problem that leads to the quality degradation of generated speech is an over-smoothing effect that eliminates some detailed structure of generated/converted acoustic features. To address this issue, we propose a synthetic-to-natural speech waveform conversion technique that uses cycle-consistent adversarial networks and which does not require any explicit assumption about speech waveform in adversarial learning. In contrast to current techniques, since our modification is performed at the waveform level, we expect that the proposed method will also make it possible to generate `vocoder-less' sounding speech even if the input speech is synthesized using a vocoder framework. The experimental results demonstrate that our proposed method can 1) alleviate the over-smoothing effect of the acoustic features despite the direct modification method used for the waveform and 2) greatly improve the naturalness of the generated speech sounds.

A different body of research has focused on quantizing and clustering network weights (Yi et al., 2008; Courbariaux et al., 2016; Rastegari et al., 2016; Deng et al., 2017; Wu et al., 2018). For successful deployment of deep neural networks on highly resource constrained devices (hearing aids, earbuds, wearables), we must simplify the types of operations and the memory/power resources required during inference. Completely avoiding inference-time floating point operations is one of the simplest ways to design networks for these highly constrained environments. By quantizing both our in-network non-linearities and our network weights, we can move to simple, compact networks without floating point operations, without multiplications, and without nonlinear function computations. Our approach allows us to explore the spectrum of possible networks, ranging from fully continuous versions down to networks with bi-level weights and activations. Our results show that quantization can be done with little or no loss of performance on both regression tasks (auto-encoding) and multi-class classification tasks (ImageNet). The memory needed to deploy our quantized networks is less than one-third of the equivalent architecture that uses floating-point operations. The activations in our networks emit only a small number of predefined, quantized values (typically 32) and all of the network's weight are drawn from a small number of unique values (typically 100-1000) found by employing a novel periodic adaptive clustering step during training. Almost all recent neural-network training algorithms rely on gradient-based learning. This has moved the research field away from using discrete-valued inference, with hard thresholds, to smooth, continuous-valued activation functions (Werbos, 1974; Rumelhart et al., 1986). Unfortunately, this causes inference to be done with floating-point operations, making it difficult to deploy on an increasinglylarge set of low-cost, limited-memory, low-power hardware in both commercial (Lane et al., 2015) and research settings (Bourzac, 2017). Avoiding all floating point operations allows the inference network to realize the power-saving gains available with fixed-point processing (Finnerty & Ratigner, 2017).

Recent studies have highlighted adversarial examples as a ubiquitous threat to different neural network models and many downstream applications. Nonetheless, as unique data properties have inspired distinct and powerful learning principles, this paper aims to explore their potentials towards mitigating adversarial inputs. In particular, our results reveal the importance of using the temporal dependency in audio data to gain discriminate power against adversarial examples. Tested on the automatic speech recognition (ASR) tasks and three recent audio adversarial attacks, we find that (i) input transformation developed from image adversarial defense provides limited robustness improvement and is subtle to advanced attacks; (ii) temporal dependency can be exploited to gain discriminative power against audio adversarial examples and is resistant to adaptive attacks considered in our experiments. Our results not only show promising means of improving the robustness of ASR systems, but also offer novel insights in exploiting domain-specific data properties to mitigate negative effects of adversarial examples. Deep Neural Networks (DNNs) have been widely adopted in a variety of machine learning applications (Krizhevsky et al., 2012; Hinton et al., 2012; Levine et al., 2016). However, recent work has demonstrated that DNNs are vulnerable to adversarial perturbations (Szegedy et al., 2014; Goodfel-low et al., 2015). An adversary can add negligible perturbations to inputs and generate adversarial examples to mislead DNNs, first found in image-based machine learning tasks (Goodfellow et al., 2015; Carlini & Wagner, 2017a; Liu et al., 2017; Chen et al., 2017b;a; Su et al., 2018). Beyond images, given the wide application of DNN-based audio recognition systems, such as Google Home and Amazon Alexa, audio adversarial examples have also been studied recently (Car-lini & Wagner, 2018; Alzantot et al., 2018; Cisse et al., 2017; Kreuk et al., 2018). Comparing between image and audio learning tasks, although their state-of-the-art DNN architectures are quite different (i.e., convolutional v.s.

Evaluation has always been a key challenge in the development of artificial intelligence (AI) based software, due to the technical complexity of the software artifact and, often, its embedding in complex sociotechnical processes. Recent advances in machine learning (ML) enabled by deep neural networks has exacerbated the challenge of evaluating such software due to the opaque nature of these ML-based artifacts. A key related issue is the (in)ability of such systems to generate useful explanations of their outputs, and we argue that the explanation and evaluation problems are closely linked. The paper models the elements of a ML-based AI system in the context of public sector decision (PSD) applications involving both artificial and human intelligence, and maps these elements against issues in both evaluation and explanation, showing how the two are related. We consider a number of common PSD application patterns in the light of our model, and identify a set of key issues connected to explanation and evaluation in each case. Finally, we propose multiple strategies to promote wider adoption of AI/ML technologies in PSD, where each is distinguished by a focus on different elements of our model, allowing PSD policy makers to adopt an approach that best fits their context and concerns.

Big data, data science, deep learning, artificial intelligence are the key words of intense hype related with a job market in full evolution, that impose to adapt the contents of our university professional trainings. Which artificial intelligence is mostly concerned by the job offers? Which methodologies and technologies should be favored in the training pprograms? Which objectives, tools and educational resources do we needed to put in place to meet these pressing needs? We answer these questions in describing the contents and operational ressources in the Data Science orientation of the speciality Applied Mathematics at INSA Toulouse. We focus on basic mathematics training (Optimization, Probability, Statistics), associated with the practical implementation of the most performing statistical learning algorithms, with the most appropriate technologies and on real examples. Considering the huge volatility of the technologies, it is imperative to train students in seft-training, this will be their technological watch tool when they will be in professional activity. This explains the structuring of the educational site https://github.com/wikistat/ into a set of tutorials. Finally, to motivate the thorough practice of these tutorials, a serious game is organized each year in the form of a prediction contest between students of Master degrees in Applied Mathematics for IA.

Deep reinforcement learning (RL) algorithms have shown an impressive ability to learn complex control policies in high-dimensional environments. However, despite the ever-increasing performance on popular benchmarks like the Arcade Learning Environment (ALE), policies learned by deep RL algorithms can struggle to generalize when evaluated in remarkably similar environments. These results are unexpected given the fact that, in supervised learning, deep neural networks often learn robust features that generalize across tasks. In this paper, we study the generalization capabilities of DQN in order to aid in understanding this mismatch between generalization in deep RL and supervised learning methods. We provide evidence suggesting that DQN overspecializes to the domain it is trained on. We then comprehensively evaluate the impact of traditional methods of regularization from supervised learning, $\ell_2$ and dropout, and of reusing learned representations to improve the generalization capabilities of DQN. We perform this study using different game modes of Atari 2600 games, a recently introduced modification for the ALE which supports slight variations of the Atari 2600 games used for benchmarking in the field. Despite regularization being largely underutilized in deep RL, we show that it can, in fact, help DQN learn more general features. These features can then be reused and fine-tuned on similar tasks, considerably improving the sample efficiency of DQN.