Concerns around China's relative weakness in basic AI infrastructure are growing, with some in the industry saying the country needs to create a popular machine learning framework or strengthen existing ones to protect against a doomsday scenario.

On October 10th, our team hosted a live webinar--Simple Distributed Deep Learning Model Inference--with Xiangrui Meng, Software Engineer at Databricks. Model inference, unlike model training, is usually embarrassingly parallel and hence simple to distribute. However, in practice, complex data scenarios and compute infrastructure often make this "simple" task hard to do from data source to sink. In this webinar, we provided a reference end-to-end pipeline for distributed deep learning model inference using the latest features from Apache Spark and Delta Lake. While the reference pipeline applies to various deep learning scenarios, we focused on image applications, and demonstrated specific pain points and proposed solutions.

Deep learning models require a very large amount of data in order to perform better than other techniques. Also, according to Tatman, just the compute of a simple image generation model in deep learning can cost around $60,000. This cost will increase with the complexity of the data models. It additionally requires expensive GPUs and hundreds of machines which will again deepen the cost to the user. Many less skilled people also find it difficult to adopt deep learning, as there is no standard theory available for learning about deep learning tools.

Can AI teach itself the laws of physics? Will classical computers soon be replaced by deep neural networks? Sure looks like it, if you've been following the news, which lately has been filled with headlines like, "A neural net solves the three-body problem 100 million times faster: Machine learning provides an entirely new way to tackle one of the classic problems of applied mathematics," and "Who needs Copernicus if you have machine learning?". The latter was described by another journalist, in an article called "AI Teaches Itself Laws of Physics," as a "monumental moment in both AI and physics," which "could be critical in solving quantum mechanics problems." The trouble is, the authors have given no compelling reason to think that they could actually do this.

Editor's Note: This deep dive companion to our high-level FAQ piece is a 30-minute read so get comfortable! You'll learn the backstory and nuances of BERT's evolution, how the algorithm works to improve human language understanding for machines and what it means for SEO and the work we do every day. If you have been keeping an eye on Twitter SEO over the past week you'll have likely noticed an uptick in the number of gifs and images featuring the character Bert (and sometimes Ernie) from Sesame Street. This is because, last week Google announced an imminent algorithmic update would be rolling out, impacting 10% of queries in search results, and also affect featured snippet results in countries where they were present; which is not trivial. The update is named Google BERT (Hence the Sesame Street connection – and the gifs). Google describes BERT as the largest change to its search system since the company introduced RankBrain, almost five years ago, and probably one of the largest changes in search ever. The news of BERT's arrival and its impending impact has caused a stir in the SEO community, along with some confusion as to what BERT does, and what it means for the industry overall. With this in mind, let's take a look at what BERT is, BERT's background, the need for BERT and the challenges it aims to resolve, the current situation (i.e. The BERT backstory How search engines learn language Problems with language learning methods How BERT improves search engine language understanding What does BERT mean for SEO? BERT is a technologically ground-breaking natural language processing model/framework which has taken the machine learning world by storm since its release as an academic research paper. The research paper is entitled BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (Devlin et al, 2018). Following paper publication Google AI Research team announced BERT as an open source contribution. A year later, Google announced a Google BERT algorithmic update rolling out in production search. Google linked the BERT algorithmic update to the BERT research paper, emphasizing BERT's importance for contextual language understanding in content and queries, and therefore intent, particularly for conversational search. BERT is described as a pre-trained deep learning natural language framework that has given state-of-the-art results on a wide variety of natural language processing tasks. Whilst in the research stages, and prior to being added to production search systems, BERT achieved state-of-the-art results on 11 different natural language processing tasks. These natural language processing tasks include, amongst others, sentiment analysis, named entity determination, textual entailment (aka next sentence prediction), semantic role labeling, text classification and coreference resolution. BERT also helps with the disambiguation of words with multiple meanings known as polysemous words, in context.

One of the effects of the ongoing trade war between the US and China is likely to be the accelerated development of what are being called "artificial intelligence chips", or AI chips for short, also sometimes referred to as AI accelerators. AI chips could play a critical role in economic growth going forward because they will inevitably feature in cars, which are becoming increasingly autonomous; smart homes, where electronic devices are becoming more intelligent; robotics, obviously; and many other technologies. AI chips, as the term suggests, refers to a new generation of microprocessors which are specifically designed to process artificial intelligence tasks faster, using less power. Obvious, you might think, but some might wonder what the difference between an AI chip and a regular chip would be when all chips of any type process zeros and ones – a typical processor, after all, is actually capable of AI tasks. Graphics-processing units are particularly good at AI-like tasks, which is why they form the basis for many of the AI chips being developed and offered today. Without getting out of our depth, while a general microprocessor is an all-purpose system, AI processors are embedded with logic gates and highly parallel calculation systems that are more suited to typical AI tasks such as image processing, machine vision, machine learning, deep learning, artificial neural networks, and so on. Maybe one could use cars as metaphors. A general microprocessor is your typical family car that might have good speed and steering capabilities.

Human emotions unfold over time, and more affective computing research has to prioritize capturing this crucial component of real-world affect. Modeling dynamic emotional stimuli requires solving the twin challenges of time-series modeling and of collecting high-quality time-series datasets. We begin by assessing the state-of-the-art in time-series emotion recognition, and we review contemporary time-series approaches in affective computing, including discriminative and generative models. We then introduce the first version of the Stanford Emotional Narratives Dataset (SENDv1): a set of rich, multimodal videos of self-paced, unscripted emotional narratives, annotated for emotional valence over time. The complex narratives and naturalistic expressions in this dataset provide a challenging test for contemporary time-series emotion recognition models. We demonstrate several baseline and state-of-the-art modeling approaches on the SEND, including a Long Short-Term Memory model and a multimodal Variational Recurrent Neural Network, which perform comparably to the human-benchmark. We end by discussing the implications for future research in time-series affective computing.

Granger causality is a widely-used criterion for analyzing interactions in large-scale networks. As most physical interactions are inherently nonlinear, we consider the problem of inferring the existence of pairwise Granger causality between nonlinearly interacting stochastic processes from their time series measurements. Our proposed approach relies on modeling the embedded nonlinearities in the measurements using a component-wise time series prediction model based on Statistical Recurrent Units (SRUs). We make a case that the network topology of Granger causal relations is directly inferrable from a structured sparse estimate of the internal parameters of the SRU networks trained to predict the processes$'$ time series measurements. We propose a variant of SRU, called economy-SRU, which, by design has considerably fewer trainable parameters, and therefore less prone to overfitting. The economy-SRU computes a low-dimensional sketch of its high-dimensional hidden state in the form of random projections to generate the feedback for its recurrent processing. Additionally, the internal weight parameters of the economy-SRU are strategically regularized in a group-wise manner to facilitate the proposed network in extracting meaningful predictive features that are highly time-localized to mimic real-world causal events. Extensive experiments are carried out to demonstrate that the proposed economy-SRU based time series prediction model outperforms the MLP, LSTM and attention-gated CNN-based time series models considered previously for inferring Granger causality.

Deep neural networks suffer from performance decay when there is domain shift between the labeled source domain and unlabeled target domain, which motivates the research on domain adaptation (DA). Conventional DA methods usually assume that the labeled data is sampled from a single source distribution. However, in practice, labeled data may be collected from multiple sources, while naive application of the single-source DA algorithms may lead to suboptimal solutions. In this paper, we propose a novel multi-source distilling domain adaptation (MDDA) network, which not only considers the different distances among multiple sources and the target, but also investigates the different similarities of the source samples to the target ones. Specifically, the proposed MDDA includes four stages: (1) pre-train the source classifiers separately using the training data from each source; (2) adversarially map the target into the feature space of each source respectively by minimizing the empirical Wasserstein distance between source and target; (3) select the source training samples that are closer to the target to fine-tune the source classifiers; and (4) classify each encoded target feature by corresponding source classifier, and aggregate different predictions using respective domain weight, which corresponds to the discrepancy between each source and target. Extensive experiments are conducted on public DA benchmarks, and the results demonstrate that the proposed MDDA significantly outperforms the state-of-the-art approaches. Our source code is released at: https://github.com/daoyuan98/MDDA.

The correspondence between residual networks and dynamical systems motivates researchers to unravel the physics of ResNets with well-developed tools in numeral methods of ODE systems. The Runge-Kutta-Fehlberg method is an adaptive time stepping that renders a good trade-off between the stability and efficiency. Can we also have an adaptive time stepping for ResNets to ensure both stability and performance? In this study, we analyze the effects of time stepping on the Euler method and ResNets. We establish a stability condition for ResNets with step sizes and weight parameters, and point out the effects of step sizes on the stability and performance. Inspired by our analyses, we develop an adaptive time stepping controller that is dependent on the parameters of the current step, and aware of previous steps. The controller is jointly optimized with the network training so that variable step sizes and evolution time can be adaptively adjusted. We conduct experiments on ImageNet and CIFAR to demonstrate the effectiveness. It is shown that our proposed method is able to improve both stability and accuracy without introducing additional overhead in inference phase.