A little over a week after the fervor surrounding Google's involvement in the Department of Defense's Project Maven, an autonomous drone program, showed signs of abating, another machine learning controversy returned to the headlines: local law enforcement deploying Amazon's Rekognition, a computer vision service with facial recognition capabilities. In a letter addressed to Amazon CEO Jeff Bezos, 19 groups of shareholders expressed concerns that Rekognition's facial recognition capabilities will be misused in ways that "violate [the] civil and human rights" of "people of color, immigrants, and civil society organizations." And they said that it set the stage for sales of the software to foreign governments and authoritarian regimes. Amazon, for its part, said in a statement that it will "suspend … customer's right to use … services [like Rekognition]" if it determines those services are being "abused." It has so far declined, however, to define the bright-line rules that would trigger a suspension.

Adam is the super star optimization algorithm of Deep Learning. Optimization algorithms aim to find optimum weights, minimize error and maximize accuracy. We find partial derivative of total error with respect to each weight and use this calculation to update weights. This is common because it works slowly but surely. In 2015, Adam optimization algorithm is raised. The name of the algorithm refers to adaptive moment estimation.

You can find links to all of the posts in the introduction, and a book based on the R series on Amazon. This blog post is a brief introduction to using the Keras deep learning framework to solve classic (shallow) machine learning problems. It presents a case study from my experience at Windfall Data, where I worked on a model to predict housing prices for hundreds of millions of properties in the US. I recently started reading "Deep Learning with R", and I've been really impressed with the support that R has for digging into deep learning. However, now that I'm porting my blog series to Python, I'll be using the Keras library directly, rather than the R wrapper.

Some key terms arise when discussing AI. Lots of people will use words such as machine learning, deep learning, algorithms and AI interchangeably. Not having a rough idea of what each term really means can have consequences when we expect results. So what does each term really mean? Firstly, I'm no expert on this topic and I'd suggest that you seek to understand from those who are - namely through an organisation like Filtered.

In this article by Maxim Lapan, the author of Deep Reinforcement Learning Hands-On,we are going to discuss about gradients in PyTorch. Gradients support in tensors is one of the major changes in PyTorch 0.4.0. In previous versions, graph tracking and gradients accumulation were done in a separate, very thin class Variable, which worked as a wrapper around the tensor and automatically performed saving of the history of computations in order to be able to backpropagate. Now gradients are a built-in tensor property, which makes the API much cleaner. Gradient was originally implemented in the Caffe toolkit and then became the de-facto standard in DL libraries.

TLDR: Neural Networks are powerful but complex and opaque tools. Using Topological Data Analysis, we can describe the functioning and learning of a convolutional neural network in a compact and understandable way. The implications of the finding are profound and can accelerate the development of a wide range of applications from self-driving everything to GDPR. Neural networks have demonstrated a great deal of success in the study of various kinds of data, including images, text, time series, and many others. One issue that restricts their applicability, however, is the fact that it is not understood in any kind of detail how they work.

Michael Nielsen provides a visual demonstration in his web book Neural Networks and Deep Learning that a 1-layer deep neural network can match any function . It is just a matter of the number of neurons to get a prediction that is arbitrarily close – the more the neurons the better the approximation. There is the Universal Approximation Theorem as well that supplies a rigorous proof of the same.But the known issues with overfitting remain and the obtained network model is only good for the range of the training data. That is, if the training data consisted only of inputs with there would be no reason to expect the obtained network model to work outside of that range. This series of posts are about obtaining network models that are unique, generic and exact.

It's no secret that identity fraud is a growing problem: A record 16.7 million US adults experienced identity fraud in 2017, marking an 8% increase from the year before, according to Javelin's 2018 Identity Fraud study. The amount of fraudulent transactions, massive data breaches, and instances of identity theft continues to rise as hackers and fraudsters become more sophisticated. ID scanning solutions have various strengths; some simply scan an ID's barcode, whereas more robust software performs forensic and biometric tests to ensure that an ID is not forged. Artificial intelligence and its subsets of machine learning and deep learning make it possible to accurately process, verify, and authenticate identities at scale. Identity documents, such as driver's licenses and passports, are scanned to test various elements of an ID, either on premises or remotely with mobile devices.

Virtual high-throughput screening provides a strategy for prioritizing compounds for physical screens. Machine learning methods offer an ancillary benefit to make molecule predictions, yet the choice of representation has been challenging when selecting algorithms. We emphasize the effects of different levels of molecule representation. Then, we introduce N-gram graph, a novel representation for a molecular graph. We demonstrate that N-gram graph is able to attain most accurate prediction with several non-deep machine learning methods on multiple tasks.

The current trend of pushing CNNs deeper with convolutions has created a pressing demand to achieve higher compression gains on CNNs where convolutions dominate the computation and parameter amount (e.g., GoogLeNet, ResNet and Wide ResNet). Further, the high energy consumption of convolutions limits its deployment on mobile devices. To this end, we proposed a simple yet effective scheme for compressing convolutions though applying k-means clustering on the weights, compression is achieved through weightsharing, by only recording K cluster centers and weight assignment indexes. We then introduced a novel spectrally relaxed k-means regularization, which tends to make hard assignments of convolutional layer weights to K learned cluster centers during retraining. We additionally propose an improved set of metrics to estimate energy consumption of CNN hardware implementations, whose estimation results are verified to be consistent with previously proposed energy estimation tool extrapolated from actual hardware measurements. We finally evaluated Deep k-Means across several CNN models in terms of both compression ratio and energy consumption reduction, observing promising results without incurring accuracy loss. The code is available at https://github.