Deep learning’s advances are the product of pattern recognition: neural networks memorize classes of things and more-or-less reliably know when they encounter them again. But most of the interesting problems in cognition aren’t classification problems at all.

Lightmatter is a startup that makes photonic chips that essentially perform calculations at the speed of light. It competes directly with GPU manufacturers and custom-built Deep Learning chips.

Paige.ai, an acronym for Pathology AI Guidance Engine, has closed $25 million in Series A funding to build a system to help understand cancer pathology. It has exclusive access to its 25 million pathology slides as well as its intellectual property related to computational pathology.

A newly published paper studied the performance of three leading face recognition systems — by Microsoft, IBM and Megvii of China — by classifying how well they could guess the gender of people with different skin tones. Microsoft’s error rate for darker-skinned women was 21 percent, while IBM’s and Megvii’s rates were nearly 35 percent. They all had error rates below 1 percent for light-skinned males.

Entropy, Cross-Entropy, and KL-Divergence are often used in Machine Learning, in particular for training classifiers. In this short video, you will understand where they come from and why we use them in ML.

Deep Reinforcement Learning has achieved remarkable success in a range of tasks, from continuous control problems in robotics to playing games like Go and Atari. The improvements seen in these domains have so far been limited to individual tasks, where a separate agent has been tuned and trained for each task. In this work, the researchers explore the challenge of training a single agent on many tasks.

A system for automatically figuring out which object is meant by a word by having a neural network decide if the word belongs to each of about 100 automatically-discovered “types” (non-exclusive categories).

(Disclosure: My own post). The academic Deep Learning research community has largely stayed away from the financial markets. In this post, I’m arguing that training Reinforcement Learning agents to trade in the financial and cryptocurrency markets can be an extremely interesting research problem.

This project is about explaining what machine learning models are doing. It currently supports explaining individual predictions for text classifiers and classifiers that act on tables or images. Check out this promo video. 

DMLab-30 is a set of environments designed for DeepMind Lab. These environments enable a researcher to develop agents for a large spectrum of interesting tasks either individually or in a multi-task setting. 28 levels are currently released.

An implementation of a Convolutional Neural Network in Google Sheets. The network classifies handwritten digits. A great way to intuitively learn how CNN filters are working.

New features include the export of stripped SavedModels from Estimators and FFT support added to XLA CPU/GPU.

The effort devoted to hand-crafting image classifiers has motivated the use of architecture search to discover them automatically. This study employs a regularized version of a popular asynchronous evolutionary algorithm which is compared to the non-regularized form and to a highly-successful reinforcement learning baseline. These models set a new state of the art for CIFAR-10 (mean test error = 2.13%) and mobile-size ImageNet (top-5 accuracy = 92.1% with 5.06M parameters), and reach the current state of the art for ImageNet (top-5 accuracy = 96.2%).

Solving a large collection of tasks using a single reinforcement learning agent with a single set of parameters. The authors developed a new distributed agent IMPALA (Importance Weighted Actor-Learner Architecture) that not only uses resources more efficiently in single-machine training but also scales to thousands of machines without sacrificing data efficiency or resource utilization. IMPALA is able to achieve better performance than previous agents with less data, and exhibits positive transfer between tasks as a result of its multi-task approach.

This survey aims to provide a comprehensive survey of the online machine learning literature through a systematic review of basic ideas and key principles and a proper categorization of different algorithms and techniques. Online learning can be classified into three major categories: (i) supervised online learning where full feedback information is always available, (ii) online learning with limited feedback, and (iii) unsupervised online learning where there is no feedback available.

This paper reviews recent studies in understanding neural-network representations and learning neural networks with interpretable/disentangled middle-layer representations. The authors focus on convolutional neural networks (CNNs), and the visualization of CNN representations, methods of diagnosing representations of pre-trained CNNs, approaches for disentangling pre-trained CNN representations, learning of CNNs with disentangled representations, and middle-to-end learning based on model interpretability.