The problem of session-based recommendation aims to predict user actions based on anonymous sessions. Previous methods model a session as a sequence and estimate user representations besides item representations to make recommendations. Though achieved promising results, they are insufficient to obtain accurate user vectors in sessions and neglect complex transitions of items. To obtain accurate item embedding and take complex transitions of items into account, we propose a novel method, i.e. Session-based Recommendation with Graph Neural Networks, SR-GNN for brevity. In the proposed method, session sequences are modeled as graph-structured data. Based on the session graph, GNN can capture complex transitions of items, which are difficult to be revealed by previous conventional sequential methods. Each session is then represented as the composition of the global preference and the current interest of that session using an attention network. Extensive experiments conducted on two real datasets show that SR-GNN evidently outperforms the state-of-the-art session-based recommendation methods consistently.

Deep learning is popular as an end-to-end framework extracting the prominent features and performing the classification also. In this paper, we extensively investigate deep networks as an alternate to feature encoding technique of low level descriptors for emotion recognition on the benchmark EmoDB dataset. Fusion performance with such obtained encoded features with other available features is also investigated. Highest performance to date in the literature is observed.

In this work, we extend the SchNet architecture by using weighted skip connections to assemble the final representation. This enables us to study the relative importance of each interaction block for property prediction. We demonstrate on both the QM9 and MD17 dataset that their relative weighting depends strongly on the chemical composition and configurational degrees of freedom of the molecules which opens the path towards a more detailed understanding of machine learning models for molecules.

Bayesian inference is known to provide a general framework for incorporating prior knowledge or specific properties into machine learning models via carefully choosing a prior distribution. In this work, we propose a new type of prior distributions for convolutional neural networks, deep weight prior, that in contrast to previously published techniques, favors empirically estimated structure of convolutional filters e.g., spatial correlations of weights. We define deep weight prior as an implicit distribution and propose a method for variational inference with such type of implicit priors. In experiments, we show that deep weight priors can improve the performance of Bayesian neural networks on several problems when training data is limited. Also, we found that initialization of weights of conventional networks with samples from deep weight prior leads to faster training.

We study the flow of information and the evolution of internal representations during deep neural network (DNN) training, aiming to demystify the compression aspect of the information bottleneck theory. The theory suggests that DNN training comprises a rapid fitting phase followed by a slower compression phase, in which the mutual information $I(X;T)$ between the input $X$ and internal representations $T$ decreases. Several papers observe compression of estimated mutual information on different DNN models, but the true $I(X;T)$ over these networks is provably either constant (discrete $X$) or infinite (continuous $X$). This work explains the discrepancy between theory and experiments, and clarifies what was actually measured by these past works. To this end, we introduce an auxiliary (noisy) DNN framework for which $I(X;T)$ is a meaningful quantity that depends on the network's parameters. This noisy framework is shown to be a good proxy for the original (deterministic) DNN both in terms of performance and the learned representations. We then develop a rigorous estimator for $I(X;T)$ in noisy DNNs and observe compression in various models. By relating $I(X;T)$ in the noisy DNN to an information-theoretic communication problem, we show that compression is driven by the progressive clustering of hidden representations of inputs from the same class. Several methods to directly monitor clustering of hidden representations, both in noisy and deterministic DNNs, are used to show that meaningful clusters form in the $T$ space. Finally, we return to the estimator of $I(X;T)$ employed in past works, and demonstrate that while it fails to capture the true (vacuous) mutual information, it does serve as a measure for clustering. This clarifies the past observations of compression and isolates the geometric clustering of hidden representations as the true phenomenon of interest.

Despite the great success of word embedding, sentence embedding remains a not-well-solved problem. In this paper, we present a supervised learning framework to exploit sentence embedding for the medical question answering task. The learning framework consists of two main parts: 1) a sentence embedding producing module, and 2) a scoring module. The former is developed with contextual self-attention and multi-scale techniques to encode a sentence into an embedding tensor. This module is shortly called Contextual self-Attention Multi-scale Sentence Embedding (CAMSE). The latter employs two scoring strategies: Semantic Matching Scoring (SMS) and Semantic Association Scoring (SAS). SMS measures similarity while SAS captures association between sentence pairs: a medical question concatenated with a candidate choice, and a piece of corresponding supportive evidence. The proposed framework is examined by two Medical Question Answering(MedicalQA) datasets which are collected from real-world applications: medical exam and clinical diagnosis based on electronic medical records (EMR). The comparison results show that our proposed framework achieved significant improvements compared to competitive baseline approaches. Additionally, a series of controlled experiments are also conducted to illustrate that the multi-scale strategy and the contextual self-attention layer play important roles for producing effective sentence embedding, and the two kinds of scoring strategies are highly complementary to each other for question answering problems.

This paper attempts to address the issues of machine learning in its current implementation. It is known that machine learning algorithms require a significant amount of data for training purposes, whereas recent developments in deep learning have increased this requirement dramatically. The performance of an algorithm depends on the quality of data and hence, algorithms are as good as the data they are trained on. Supervised learning is developed based on human learning processes by analysing named (i.e. The fact is, that training algorithms the same way we learn ourselves, comes with limitations. This paper discusses the issues around applying a humanlike approach to train algorithms and the implications of this approach when using limited data. Several current studies involving non-data-driven algorithms and natural examples are also discussed and certain alternative approaches are suggested. Keywords: machine leaning, deep learning, computer vision, action recognition, big data, data-driven, synthetic, simulation 1. Introduction The human learning process involves the gradual brain development as we grow from infancy towards adulthood. This capability though, may not always provide accurate results and can often lead to classification errors e.g a clementine may be confused with an orange. The primitive humans' ability to process learning involved a type of a rather incomplete "unsupervised learning", as they observed objects but had not developed a language system to annotate them. The cognitive process where the next generations classified or named objects, is effectively what unsupervised machine learning does by clustering. The unsupervised approach can separate classes, nevertheless, an algorithm has no actual understanding of the action or object that is clustered. On the other hand, in supervised approach, every object or action has an annotation, hence the training algorithm uses an identifier of this particular action or object.

The overarching problem in artificial intelligence (AI) is that we do not understand the intelligence process well enough to enable the development of adequate computational models. Much work has been done in AI over the years at lower levels, but a big part of what has been missing involves the high level, abstract, general nature of intelligence. We address this gap by developing a model for general intelligence. To accomplish this, we focus on three basic aspects of intelligence. First, we must realize the general order and nature of intelligence at a high level. Second, we must come to know what these realizations mean with respect to the overall intelligence process. Third, we must describe these realizations as clearly as possible. We propose a hierarchical model to help capture and exploit the order within intelligence. The underlying order involves patterns of signals that become organized, stored and activated in space and time. These patterns can be described using a simple, general hierarchy, with physical signals at the lowest level, information in the middle, and abstract signal representations at the top. This high level perspective provides a big picture that literally helps us see the intelligence process, thereby enabling fundamental realizations, a better understanding and clear descriptions of the intelligence process. The resulting model can be used to support all kinds of information processing across multiple levels of abstraction. As computer technology improves, and as cooperation increases between humans and computers, people will become more efficient and more productive in performing their information processing tasks.

Citizen science projects are successful at gathering rich datasets for various applications. However, the data collected by citizen scientists are often biased --- in particular, aligned more with the citizens' preferences than with scientific objectives. We propose the Shift Compensation Network (SCN), an end-to-end learning scheme which learns the shift from the scientific objectives to the biased data while compensating for the shift by re-weighting the training data. Applied to bird observational data from the citizen science project eBird, we demonstrate how SCN quantifies the data distribution shift and outperforms supervised learning models that do not address the data bias. Compared with competing models in the context of covariate shift, we further demonstrate the advantage of SCN in both its effectiveness and its capability of handling massive high-dimensional data.

Five years ago, the coders at DeepMind, a London-based artificial intelligence company, watched excitedly as an AI taught itself to play a classic arcade game. They'd used the hot technique of the day, deep learning, on a seemingly whimsical task: mastering Breakout,1 the Atari game in which you bounce a ball at a wall of bricks, trying to make each one vanish. Deep learning is self-education for machines; you feed an AI huge amounts of data, and eventually it begins to discern patterns all by itself. In this case, the data was the activity on the screen--blocky pixels representing the bricks, the ball, and the player's paddle. The DeepMind AI, a so-called neural network made up of layered algorithms, wasn't programmed with any knowledge about how Breakout works, its rules, its goals, or even how to play it.