Despite the recent success of deep transfer learning approaches in NLP, there is a lack of quantitative studies demonstrating the gains these models offer in low-shot text classification tasks over existing paradigms. Deep transfer learning approaches such as BERT and ULMFiT demonstrate that they can beat state-of-the-art results on larger datasets, however when one has only 100-1000 labelled examples per class, the choice of approach is less clear, with classical machine learning and deep transfer learning representing valid options. This paper compares the current best transfer learning approach with top classical machine learning approaches on a trinary sentiment classification task to assess the best paradigm. We find that BERT, representing the best of deep transfer learning, is the best performing approach, outperforming top classical machine learning algorithms by 9.7% on average when trained with 100 examples per class, narrowing to 1.8% at 1000 labels per class. We also show the robustness of deep transfer learning in moving across domains, where the maximum loss in accuracy is only 0.7% in similar domain tasks and 3.2% cross domain, compared to classical machine learning which loses up to 20.6%.

Bayesian inference was once a gold standard for learning with neural networks, providing accurate full predictive distributions and well calibrated uncertainty. However, scaling Bayesian inference techniques to deep neural networks is challenging due to the high dimensionality of the parameter space. In this paper, we construct low-dimensional subspaces of parameter space, such as the first principal components of the stochastic gradient descent (SGD) trajectory, which contain diverse sets of high performing models. In these subspaces, we are able to apply elliptical slice sampling and variational inference, which struggle in the full parameter space. We show that Bayesian model averaging over the induced posterior in these subspaces produces accurate predictions and well calibrated predictive uncertainty for both regression and image classification.

We investigate the difficulties of training sparse neural networks and make new observations about optimization dynamics and the energy landscape within the sparse regime. Recent work of \citep{Gale2019, Liu2018} has shown that sparse ResNet-50 architectures trained on ImageNet-2012 dataset converge to solutions that are significantly worse than those found by pruning. We show that, despite the failure of optimizers, there is a linear path with a monotonically decreasing objective from the initialization to the "good" solution. Additionally, our attempts to find a decreasing objective path from "bad" solutions to the "good" ones in the sparse subspace fail. However, if we allow the path to traverse the dense subspace, then we consistently find a path between two solutions. These findings suggest traversing extra dimensions may be needed to escape stationary points found in the sparse subspace.

Recently, artificial intelligence, especially machine learning has demonstrated remarkable performances in many tasks, from image processing to natural language processing, especially with the advent of deep learning. Along with research progress, machine learning has encroached into many different fields and disciplines. Some of them, such as the medical field, require high level of accountability, and thus transparency, which means we need to be able to explain machine decisions, predictions and justify their reliability. This requires greater interpretability, which often means we need to understand the mechanism underlying the algorithms. Unfortunately, the black-box nature of the deep learning is still unresolved, and many machine decisions are still poorly understood. We provide a review on interpretabilities suggested by different research works and categorize them, with the intention of providing alternative perspective that is hopefully more tractable for future adoption of interpretability standard. We explore further into interpretability in the medical field, illustrating the complexity of interpretability issue.

As artificial intelligence (AI) research and development continues to strengthen, there have been some incredibly intriguing projects where machines battled man in tasks that were once thought the realm of humans. While not all were 100% successful, AI researchers and technology companies learned a lot about how to continue forward momentum as well as what a future might look like when machines and humans work alongside one another. Here are some of the highlights from when artificial intelligence battled humans. World Champion chess player Garry Kasparov competed against artificial intelligence twice. In the first chess match-up between machine (IBM Deep Blue) and man (Kasparov) in 1996 Kasparov won.

A deep-learning algorithm has been developed which can solve the Rubik's cube faster than any human can. It never fails to complete the puzzle, with a 100 per cent success rate and managing it in around 20 moves. Humans can beat the AI's mark of 18 seconds, the world record is around four seconds, but it is far more inefficient and people often require around 50 moves. It was created by University of California Irvine and can be tried out here. Given an unsolved cube, the machine must decide whether a specific move is an improvement on the existing configuration.

Clara Medical Imaging is a collection of developer toolkits built on NVIDIA's compute platform aimed at accelerating compute, artificial intelligence, and advanced visualization. Medical imaging industry is being transformed. A decade ago, the earliest applications to take advantage of GPU computing were image & signal processing applications. Today, GPUs are found in almost all imaging modalities, including CT, MRI, X-ray, and Ultrasound bringing more compute capabilities to the edge devices. Deep Learning research in Medical Imaging is also booming with more efficient and improved approaches being developed to enable AI-assisted workflows.Today, most of this AI research is being done in isolation and with limited datasets which may lead to overly simplified models.

About the author Andrew Macleod is the director of automotive marketing at Siemens, focusing on the Mentor product suite. He has more than 15 years of experience in the automotive software and semiconductor industry, with expertise in new product development and introduction, automotive integrated circuit product management and global strategy, including a focus on the Chinese auto industry. He earned a 1st class honors engineering degree from the University of Paisley in the UK and lives in Austin, Texas. Follow him on Twitter @AndyMacleod_MG.

The "little black dress" has been considered a staple in women's fashion since the designer Coco Chanel popularized it in the 1920s. Since then, it's seen many iterations, most recently, by machine-learning software developed by two recent MIT graduates, called Glitch. Pinar Yanardag and Emily Salvador met at MIT while taking a course called "How to Generate (Almost) Anything," which encouraged students to use deep learning software for creative projects. In that course, they dabbled with creating AI-generated art, perfume and jewelry, and were inspired to start Glitch, a new clothing company that sells pieces designed by AI. "The'little black dress' is considered an essential item that should be in any woman's wardrobe," said Yanardag. "Sooner or later, AI is going to be an essential tool for any person in computing, so we thought the'little black dress' was a good place to start."

Machine-learning technology has beaten humans at games of chess and Go to worldwide fanfare. A demonstration of its eerily lifelike prowess in making phone calls to unsuspecting people went viral. But a less-noticed win for DeepMind, the artificial-intelligence arm of Google's parent Alphabet Inc., at a biennial biology conference could upend how drugmakers find and develop new medicines. It could also dial up pressure on the world's largest pharmaceutical companies to prepare for a technological arms race. Already, a new breed of upstarts are jumping into the fray.