In this work, we introduce a novel probabilistic representation of deep learning, which provides an explicit explanation for the Deep Neural Networks (DNNs) in three aspects: (i) neurons define the energy of a Gibbs distribution; (ii) the hidden layers of DNNs formulate Gibbs distributions; and (iii) the whole architecture of DNNs can be interpreted as a Bayesian neural network. Based on the proposed probabilistic representation, we investigate two fundamental properties of deep learning: hierarchy and generalization. First, we explicitly formulate the hierarchy property from the Bayesian perspective, namely that some hidden layers formulate a prior distribution and the remaining layers formulate a likelihood distribution. Second, we demonstrate that DNNs have an explicit regularization by learning a prior distribution and the learning algorithm is one reason for decreasing the generalization ability of DNNs. Moreover, we clarify two empirical phenomena of DNNs that cannot be explained by traditional theories of generalization. Simulation results validate the proposed probabilistic representation and the insights into these properties of deep learning based on a synthetic dataset.

Underlying cause of death coding from death certificates is a process that is nowadays undertaken mostly by humans with a potential assistance from expert systems such as the Iris software. It is as a consequence an expensive process that can in addition suffer from geospatial discrepancies, thus severely impairing the comparability of death statistics at the international level. The recent advances in artificial intelligence, specifically the raise of deep learning methods, has enabled computers to make efficient decisions on a number of complex problem that were typically considered as out of reach without human assistance. They however require a considerable amount of data to learn from, which is typically their main limiting factor. However, the C\'epiDc stores an exhaustive database of death certificate at the French national scale, amounting to several millions training example available for the machine learning practitioner. This article presents a deep learning based tool for automated coding of the underlying cause of death from the data contained in death certificates with 97.8% accuracy, a substantial achievement compared to the Iris software and its 75% accuracy assessed on the same test examples. Such an improvement opens a whole field of new applications, from nosologist-level batch automated coding to international and temporal harmonization of cause of death statistics.

We present an analysis of predictive uncertainty based out-of-distribution detection for different approaches to estimate various models' epistemic uncertainty and contrast it with extreme value theory based open set recognition. While the former alone does not seem to be enough to overcome this challenge, we demonstrate that uncertainty goes hand in hand with the latter method. This seems to be particularly reflected in a generative model approach, where we show that posterior based open set recognition outperforms discriminative models and predictive uncertainty based outlier rejection, raising the question of whether classifiers need to be generative in order to know what they have not seen.

Outfittery is an online personalized styling service targeted at men. We have hundreds of stylists who create thousands of bespoke outfits for our customers every day. A critical challenge faced by our stylists when creating these outfits is selecting an appropriate item of clothing that makes sense in the context of the outfit being created, otherwise known as style fit. Another significant challenge is knowing if the item is relevant to the customer based on their tastes, physical attributes and price sensitivity. At Outfittery we leverage machine learning extensively and combine it with human domain expertise to tackle these challenges. We do this by surfacing relevant items of clothing during the outfit building process based on what our stylist is doing and what the preferences of our customer are. In this paper we describe one way in which we help our stylists to tackle style fit for a particular item of clothing and its relevance to an outfit. A thorough qualitative and quantitative evaluation highlights the method's ability to recommend fashion items by style fit.

We propose an iteration-free source separation algorithm based on Winner-Take-All (WTA) hash codes, which is a faster, yet accurate alternative to a complex machine learning model for single-channel source separation in a resource-constrained environment. We first generate random permutations with WTA hashing to encode the shape of the multidimensional audio spectrum to a reduced bitstring representation. A nearest neighbor search on the hash codes of an incoming noisy spectrum as the query string results in the closest matches among the hashed mixture spectra. Using the indices of the matching frames, we obtain the corresponding ideal binary mask vectors for denoising. Since both the training data and the search operation are bitwise, the procedure can be done efficiently in hardware implementations. Experimental results show that the WTA hash codes are discriminant and provide an affordable dictionary search mechanism that leads to a competent performance compared to a comprehensive model and oracle masking.

Embedding layer is commonly used to map discrete symbols into continuous embedding vectors that reflect their semantic meanings. As the number of symbols increase, the number of embedding parameter, as well as their size, increase linearly and become problematically large. In this work, we aim to reduce the size of embedding layer via learning discrete codes and composing embedding vectors from the codes. More specifically, we propose a differentiable product quantization framework with two instantiations, which can serve as an efficient drop-in replacement for existing embedding layer. Empirically, we evaluate the proposed method on three different language tasks, and show that the proposed method enables end-to-end training of embedding compression that achieves significant compression ratios (14-238$\times$) at almost no performance cost (sometimes even better).

Background Based Conversations (BBCs) have been introduced to help conversational systems avoid generating overly generic responses. In a BBC the conversation is grounded in a knowledge source. A key challenge in BBCs is Knowledge Selection (KS): given a conversation context, try to find the appropriate background knowledge (a text fragment containing related facts or comments, etc.) based on which to generate the next response. Previous work addresses KS by employing attention and/or pointer mechanisms. These mechanisms use a local perspective, i.e., they select a token at a time based solely on the current decoding state. We argue for the adoption of a global perspective, i.e., pre-selecting some text fragments from the background knowledge that could help determine the topic of the next response. We enhance KS in BBCs by introducing a Global-to-Local Knowledge Selection (GLKS) mechanism. Given a conversation context and background knowledge, we first learn a topic transition vector to encode the most likely text fragments to be used in the next response, which is then used to guide the local KS at each decoding timestamp. In order to effectively learn the topic transition vector, we propose a distantly supervised learning schema. Experimental results show that the GLKS model significantly outperforms state-of-the-art methods in terms of both automatic and human evaluations. More importantly, GLKS achieves this without requiring any extra annotations, which demonstrates its high scalability.

EEG signals measure the neuronal activities on different brain regions via electrodes. Many existing studies on EEG-based emotion recognition do not exploit the topological structure of EEG signals. In this paper, we propose a regularized graph neural network (RGNN) for EEG-based emotion recognition, which is biologically supported and captures both local and global inter-channel relations. Specifically, we model the inter-channel relations in EEG signals via an adjacency matrix in our graph neural network where the connection and sparseness of the adjacency matrix are supported by the neurosicience theories of human brain organization. In addition, we propose two regularizers, namely node-wise domain adversarial training (NodeDAT) and emotion-aware distribution learning (EmotionDL), to improve the robustness of our model against cross-subject EEG variations and noisy labels, respectively. To thoroughly evaluate our model, we conduct extensive experiments in both subject-dependent and subject-independent classification settings on two public datasets: SEED and SEED-IV. Our model obtains better performance than competitive baselines such as SVM, DBN, DGCNN, BiDANN, and the state-of-the-art BiHDM in most experimental settings . Our model analysis demonstrates that the proposed biologically supported adjacency matrix and two regularizers contribute consistent and significant gain to the performance. Investigations on the neuronal activities reveal that pre-frontal, parietal and occipital regions may be the most informative regions for emotion recognition, which is consistent with relevant prior studies. In addition, experimental results suggest that global inter-channel relations between the left and right hemispheres are important for emotion recognition and local inter-channel relations between (FP1, AF3), (F6, F8) and (FP2, AF4) may also provide useful information.

In his writings, Gary Marcus is clear about two things: Artificial intelligence is an extremely promising technology that, if used in the right way, could significantly improve practices in health care and other industries. But right now, Marcus says, AI is getting off track, with potentially severe consequences for society and the field itself. That viewpoint makes Marcus -- a tech entrepreneur, author, and psychology professor at New York University -- a controversial figure in the world of artificial intelligence. He is among a few prominent scientists voicing skepticism about the dominance of deep learning, a type of AI architecture whose use has exploded in medicine and other fields. Unlock this article by subscribing to STAT Plus and enjoy your first 30 days free!

Even the most experienced doctors can't catch every tiny polyp during an endoscopy, a screening of the digestive system. But even in routine exams, the stakes are high -- missing an early warning sign of cancer can lead to delayed diagnosis and treatment, lowering a patient's chances for recovery. To cut down on the rate of missed precancerous lesions, one Japanese endoscopist is turning to AI. His startup, AIM (short for AI Medical Service), is building a GPU-powered AI system that will analyze endoscopy video feeds in real time, spotting lesions and helping doctors identify which are cancerous or at risk of becoming so. AI screening could also help clinicians manage a demanding workload: Japanese endoscopists must check more than 3,000 medical images a day, on average.