This paper proposes a method for machine learning from unlabeled data in the form of a time-series. The mapping that is learned is shown to extract slowly evolving information that would be useful for control applications, while efficiently filtering out unwanted, higher-frequency noise. The method consists of training a feedforward artificial neural network with backpropagation using two opposing objectives. The first of these is to minimize the squared changes in activations between time steps of each unit in the network. This "temporal smoothing" has the effect of correlating inputs that occur close in time with outputs that are close in the L2-norm. The second objective is to maximize the log determinant of the covariance matrix of activations in each layer of the network. This objective ensures that information from each layer is passed through to the next. This second objective acts as a balance to the first, which on its own would result in a network with all input weights equal to zero.

We propose an end-to-end deep learning learning model for graph classification and representation learning that is invariant to permutation of the nodes of the input graphs. We address the challenge of learning a fixed size graph representation for graphs of varying dimensions through a differentiable node attention pooling mechanism. In addition to a theoretical proof of its invariance to permutation, we provide empirical evidence demonstrating the statistically significant gain in accuracy when faced with an isomorphic graph classification task given only a small number of training examples. We analyse the effect of four different matrices to facilitate the local message passing mechanism by which graph convolutions are performed vs. a matrix parametrised by a learned parameter pair able to transition smoothly between the former. Finally, we show that our model achieves competitive classification performance with existing techniques on a set of molecule datasets.

Recently, Geometric Deep Learning (GDL) has been introduced as a novel and versatile framework for computer-aided disease classification. GDL uses patient meta-information such as age and gender to model patient cohort relations in a graph structure. Concepts from graph signal processing are leveraged to learn the optimal mapping of multi-modal features, e.g. from images to disease classes. Related studies so far have considered image features that are extracted in a pre-processing step. We hypothesize that such an approach prevents the network from optimizing feature representations towards achieving the best performance in the graph network. We propose a new network architecture that exploits an inductive end-to-end learning approach for disease classification, where filters from both the CNN and the graph are trained jointly. We validate this architecture against state-of-the-art inductive graph networks and demonstrate significantly improved classification scores on a modified MNIST toy dataset, as well as comparable classification results with higher stability on a chest X-ray image dataset. Additionally, we explain how the structural information of the graph affects both the image filters and the feature learning.

In this report we review memory-based meta-learning as a tool for building sample-efficient strategies that learn from past experience to adapt to any task within a target class. Our goal is to equip the reader with the conceptual foundations of this tool for building new, scalable agents that operate on broad domains. To do so, we present basic algorithmic templates for building near-optimal predictors and reinforcement learners which behave as if they had a probabilistic model that allowed them to efficiently exploit task structure. Furthermore, we recast memory-based meta-learning within a Bayesian framework, showing that the meta-learned strategies are near-optimal because they amortize Bayes-filtered data, where the adaptation is implemented in the memory dynamics as a state-machine of sufficient statistics. Essentially, memory-based meta-learning translates the hard problem of probabilistic sequential inference into a regression problem.

The Adam algorithm has become extremely popular for large-scale machine learning. Under convexity condition, it has been proved to enjoy a data-dependant $O(\sqrt{T})$ regret bound where $T$ is the time horizon. However, whether strong convexity can be utilized to further improve the performance remains an open problem. In this paper, we give an affirmative answer by developing a variant of Adam (referred to as SAdam) which achieves a data-dependant $O(\log T)$ regret bound for strongly convex functions. The essential idea is to maintain a faster decaying yet under controlled step size for exploiting strong convexity. In addition, under a special configuration of hyperparameters, our SAdam reduces to SC-RMSprop, a recently proposed variant of RMSprop for strongly convex functions, for which we provide the first data-dependent logarithmic regret bound. Empirical results on optimizing strongly convex functions and training deep networks demonstrate the effectiveness of our method.

Aito, a Helsinki-based machine learning startup that is developing "predictive database" technology, has raised €1 million in pre-seed funding, including from Nokia Chairman Risto Siilasmaa (via his investment company First Fellow Partners). Others backing the round are Hermitage, UMO Capital, together with funding from Business Finland. Aiming to replace current machine learning tools that have a steep learning curve and generate only single-purpose models, Aito has built a predictive database for developers. Specifically, it lets users search existing information, make predictions, and find hidden correlations. Crucially, the results are said to fully explainable and the tech can be integrated into existing software as easy as integrating an SQL query.

Does your bank help you reflect on spending habits and help you make wiser decisions? Did it detect that it wasn't you who made that last purchase? By combining the powers of big data, artificial intelligence, and machine learning, banks can do much more than keep your money and pay you a nominal interest rate. "Banking is necessary, banks are not." That's something Bill Gates said way back in the 90s, but the true weight of that short sentence is being felt today.

Check out part one of this series for what the next five to fifteen years looks like in AI. In part two we get super sci-fi and see if our crystal ball can reach 50 years into the future. Once dumb objects have woken up. Your shirt is babbling away with your shades and having a conversation with your girlfriend's pearl earrings when she's traveling to give a talk in Brazil. Everything from our houses, to weapons, to planes, trains and automobiles, to roads, clothes, jewelry, headphones, glasses, and eye contacts are wild with thoughts. The dynamic new algorithms that pushed us past deep learning and powered the fourth wave of the intelligence revolution sprang from world wide efforts to map every single neuron and connection in the human brain. Eventually the processors and biotechnology caught up with our ambitions and scientists succeeded beyond our wildest expectations.

AstraZeneca announced a long-term collaboration deal with BenevolentAI, a UK-based company focused on combining computational medicine and advanced artificial intelligence. The two companies will focus on using AI and machine learning to discover and develop new drugs for chronic kidney disease (CKD) and idiopathic pulmonary fibrosis (IPF). "The vast amount of data available to research scientists is growing exponentially each year," stated Mene Pangalo, AstraZeneca's executive vice president and president BioPharmaceuticals R&D. "By combining AstraZeneca's disease area expertise and large, diverse datasets with BenevolentAI's leading AI and machine learning capabilities, we can unlock the potential of this wealth of data to improve our understanding of complex disease biology and identify new targets that could treat debilitating diseases." No financial details were disclosed.

Mental illnesses affect 15.5% of the global population. They're the leading cause of disability-adjusted life years, accounting for 37% of healthy years lost from non-communicable diseases. Even worse, mental health conditions are on the rise, with some estimates putting the costs to the global economy up to $16 trillion between 2010 and 2030 if global healthcare systems can't tackle the problem. More than half of all mental illnesses remain untreated, leading to worsening mental conditions, chronic physical health problems, job stability issues, homelessness, trauma, and suicide. Understaffed and underfunded healthcare systems are unable to cope with this mental health crisis, and that's where technology can help alleviate the burden of mental illness on society by helping clinicians diagnose and treat mental conditions while giving patients tools to better manage their conditions on a daily basis.