Modeled after the human mind, AI thinks like we do by learning and adapting. AI systems take in information in the form of data, process it and store it for future reference. To give you a rough idea of how AI works -- think of a kid that touches a hot stove. His/her brain will register the pain and make a note not to repeat again. AI by itself is the outermost layer of four concentric circles.

The identification of patients with aggressive cancer who require immediate therapy is a health challenge in low- and middle-income countries. Limited pathology resources, high healthcare costs and large caseloads call for the development of advanced stand-alone diagnostics. Here, we report and validate an automated, low-cost point-of-care device for the molecular diagnosis of aggressive lymphomas. The device uses contrast-enhanced microholography and a deep learning algorithm to directly analyse percutaneously obtained fine-needle aspirates. We show the feasibility and high accuracy of the device in cells, as well as the prospective validation of the results in 40 patients clinically referred for image-guided aspiration of nodal mass lesions suspicious of lymphoma.

Getting into Machine Learning and AI is not an easy task. Many aspiring professionals and enthusiasts find it hard to establish a proper path into the field, given the enormous amount of resources available today. The field is evolving constantly and it is crucial that we keep up with the pace of this rapid development. In order to cope with this overwhelming speed of evolution and innovation, a good way to stay updated and knowledgeable on the advances of ML, is to engage with the community by contributing to the many open-source projects and tools that are used daily by advanced professionals. Here we update the information and examine the trends since our previous post Top 20 Python Machine Learning Open Source Projects (Nov 2016).

Recently we saw a new "Master algorithm" that could be used to create the first generation of super intelligent machines, and now a team of researchers from Maryland, USA, announced this week that they've invented a general Artificial Intelligence (AI) way for machines to identify and process 3D images that doesn't require humans to go through the tedium of inputting specific information that accounts for each and every instance, scenario, difference, change and category that could crop up, and they claim it's a world first, even though it follows on from a not too dissimilar breakthrough from Google DeepMind whose own platform, Alpha Zero, recently taught itself a mix of board games including chess to a grand master level, in just four hours. All that said though it's important to say here that while doing away with huge volumes of training data used to train today's AI's is a staggering breakthrough of potential epic proportions this announcement doesn't yet herald the emergence of fabled Artificial General Intelligence (AGI) which is still a little way off and which Google Deepmind published an architecture for last year. However, this new development, albeit a distant cousin of AGI for now, will no doubt have an influence on the speed at which the first AGI's now emerge. This is actually a huge deal for the technology sector, and it's a massive step for the case of general AI versus specific AI, and once fine tuned, this development will have the power to shape and change how everyone, from police and intelligence officials, to retail marketers and medical professionals, go about their daily business. Currently, neural networks, defined as those computing systems that are aimed at mimicking how humans think and make decisions, are only as good as the information that's inputted.

Learning-based pattern classifiers, including deep networks, have shown impressive performance in several application domains, ranging from computer vision to cybersecurity. However, it has also been shown that adversarial input perturbations carefully crafted either at training or at test time can easily subvert their predictions. The vulnerability of machine learning to such wild patterns (also referred to as adversarial examples), along with the design of suitable countermeasures, have been investigated in the research field of adversarial machine learning. In this work, we provide a thorough overview of the evolution of this research area over the last ten years and beyond, starting from pioneering, earlier work on the security of non-deep learning algorithms up to more recent work aimed to understand the security properties of deep learning algorithms, in the context of computer vision and cybersecurity tasks. We report interesting connections between these apparently-different lines of work, highlighting common misconceptions related to the security evaluation of machine-learning algorithms. We review the main threat models and attacks defined to this end, and discuss the main limitations of current work, along with the corresponding future challenges towards the design of more secure learning algorithms.

The article focuses on determining the predictive uncertainty of a model on the example of atrial fibrillation detection problem by a single-lead ECG signal. To this end, the model predicts parameters of the beta distribution over class probabilities instead of these probabilities themselves. It was shown that the described approach allows to detect atypical recordings and significantly improve the quality of the algorithm on confident predictions.

In typical machine learning tasks and applications, it is necessary to obtain or create large labeled datasets in order to to achieve high performance. Unfortunately, large labeled datasets are not always available and can be expensive to source, creating a bottleneck towards more widely applicable machine learning. The paradigm of weak supervision offers an alternative that allows for integration of domain-specific knowledge by enforcing constraints that a correct solution to the learning problem will obey over the output space. In this work, we explore the application of this paradigm to 2-D physical systems governed by non-linear differential equations. We demonstrate that knowledge of the partial differential equations governing a system can be encoded into the loss function of a neural network via an appropriately chosen convolutional kernel. We demonstrate this by showing that the steady-state solution to the 2-D heat equation can be learned directly from initial conditions by a convolutional neural network, in the absence of labeled training data. We also extend recent work in the progressive growing of fully convolutional networks to achieve high accuracy (< 1.5% error) at multiple scales of the heat-flow problem, including at the very large scale (1024x1024). Finally, we demonstrate that this method can be used to speed up exact calculation of the solution to the differential equations via finite difference.

In this paper we introduce Curriculum GANs, a curriculum learning strategy for training Generative Adversarial Networks that increases the strength of the discriminator over the course of training, thereby making the learning task progressively more difficult for the generator. We demonstrate that this strategy is key to obtaining state-of-the-art results in image generation. We also show evidence that this strategy may be broadly applicable to improving GAN training in other data modalities.

Reinforcement Learning (RL) is a learning paradigm concerned with learning to control a system so as to maximize an objective over the long term. This approach to learning has received immense interest in recent times and success manifests itself in the form of human-level performance on games like \textit{Go}. While RL is emerging as a practical component in real-life systems, most successes have been in Single Agent domains. This report will instead specifically focus on challenges that are unique to Multi-Agent Systems interacting in mixed cooperative and competitive environments. The report concludes with advances in the paradigm of training Multi-Agent Systems called \textit{Decentralized Actor, Centralized Critic}, based on an extension of MDPs called \textit{Decentralized Partially Observable MDP}s, which has seen a renewed interest lately.