Synchronous mini-batch SGD is state-of-the-art for large-scale distributed machine learning. However, in practice, its convergence is bottlenecked by slow communication rounds between worker nodes. A natural solution to reduce communication is to use the \emph{`local-SGD'} model in which the workers train their model independently and synchronize every once in a while. This algorithm improves the computation-communication trade-off but its convergence is not understood very well. We propose a non-asymptotic error analysis, which enables comparison to \emph{one-shot averaging} i.e., a single communication round among independent workers, and \emph{mini-batch averaging} i.e., communicating at every step. We also provide adaptive lower bounds on the communication frequency for large step-sizes ($ t^{-\alpha} $, $ \alpha\in (1/2 , 1 ) $) and show that \emph{Local-SGD} reduces communication by a factor of $O\Big(\frac{\sqrt{T}}{P^{3/2}}\Big)$, with $T$ the total number of gradients and $P$ machines.

Graph based clustering is one of the major clustering methods. Most of it work in three separate steps: similarity graph construction, clustering label relaxing and label discretization with k-means. Such common practice has three disadvantages: 1) the predefined similarity graph is often fixed and may not be optimal for the subsequent clustering. 2) the relaxing process of cluster labels may cause significant information loss. 3) label discretization may deviate from the real clustering result since k-means is sensitive to the initialization of cluster centroids. To tackle these problems, in this paper, we propose an effective discrete optimal graph clustering (DOGC) framework. A structured similarity graph that is theoretically optimal for clustering performance is adaptively learned with a guidance of reasonable rank constraint. Besides, to avoid the information loss, we explicitly enforce a discrete transformation on the intermediate continuous label, which derives a tractable optimization problem with discrete solution. Further, to compensate the unreliability of the learned labels and enhance the clustering accuracy, we design an adaptive robust module that learns prediction function for the unseen data based on the learned discrete cluster labels. Finally, an iterative optimization strategy guaranteed with convergence is developed to directly solve the clustering results. Extensive experiments conducted on both real and synthetic datasets demonstrate the superiority of our proposed methods compared with several state-of-the-art clustering approaches.

Deep network compression has been achieved notable progress via knowledge distillation, where a teacher-student learning manner is adopted by using predetermined loss. Recently, more focuses have been transferred to employ the adversarial training to minimize the discrepancy between distributions of output from two networks. However, they always emphasize on result-oriented learning while neglecting the scheme of process-oriented learning, leading to the loss of rich information contained in the whole network pipeline. Inspired by the assumption that, the small network can not perfectly mimic a large one due to the huge gap of network scale, we propose a knowledge transfer method, involving effective intermediate supervision, under the adversarial training framework to learn the student network. To achieve powerful but highly compact intermediate information representation, the squeezed knowledge is realized by task-driven attention mechanism. Then, the transferred knowledge from teacher network could accommodate the size of student network. As a result, the proposed method integrates merits from both process-oriented and result-oriented learning. Extensive experimental results on three typical benchmark datasets, i.e., CIFAR-10, CIFAR-100, and ImageNet, demonstrate that our method achieves highly superior performances against other state-of-the-art methods.

In recent years, there has been an exponential growth in the number of complex documents and texts that require a deeper understanding of machine learning methods to be able to accurately classify texts in many applications. Many machine learning approaches have achieved surpassing results in natural language processing. The success of these learning algorithms relies on their capacity to understand complex models and non-linear relationships within data. However, finding suitable structures, architectures, and techniques for text classification is a challenge for researchers. In this paper, a brief overview of text classification algorithms is discussed. This overview covers different text feature extractions, dimensionality reduction methods, existing algorithms and techniques, and evaluations methods. Finally, the limitations of each technique and their application in the real-world problem are discussed.

In Japan's time-scarce, results-oriented society, people no longer feel they can find a life partner through traditional dating methods, and are instead turning to internet matchmaking options to better their chances of meeting a compatible companion. Rather than visiting a dating agency, attending matchmaking parties or actually finding a partner the old-fashioned way through "a chance encounter," people are peering into their screens in hopes that artificial intelligence will help them find a match made in heaven. The companies are not focused on delivering a solely digital date, however, as some also host events where prospective partners can meet in person to see if the profile picture meets reality. Makoto Yamada, 30, who works in the western Tokyo suburb of Tachikawa, married Sayaka, 33, a university research fellow, in June last year after meeting through the Pairs online matching service run by Tokyo-based Eureka Inc. Both learned of the matchmaking service through social media ads and signed up without giving it a second thought.

The government will set guidelines by the end of March next year for preventing technology leaks from universities that conduct research with foreign firms, sources close to the matter said Wednesday. The move comes as the United States and China grow cautious about advanced technologies such as artificial intelligence being converted for military use. While Japan already regulates the disclosure of sensitive technologies and products by the nation's state organizations and companies to overseas firms under a foreign exchange and foreign trade law, university laboratories have been managing infrequent arrangements on their own, leading some experts to voice concerns about the risk of information leaks. The envisioned guidelines would require universities and other research institutions to set regulations on joint projects involving foreign entities. They will be based on the comprehensive innovation strategy adopted by the Cabinet in 2018 aimed at promoting university research on AI, biotechnology and other leading technologies.

Hitachi Ltd. said Wednesday it has reached a deal to buy U.S. assembly robot-maker JR Automation Technologies LLC for $1.43 billion (¥160 billion) to strengthen its factory automation business in the American market. Hitachi said it agreed Tuesday to buy the manufacturer from private equity firm Crestview Partners, which holds a 93 percent stake in the company. The Japanese manufacturer will acquire all shares in the Michigan-based technology provider by the end of this year. JR Automation Technologies was founded in 1980 and has strengths in automated manufacturing and technology solutions. The company has about 2,000 employees at 23 manufacturing facilities in North America, Europe and Asia, it said.

Facial recognition technology is being deployed in airports, security cameras and in our phones. Now, Tokyo is using facial recognition in an unexpected way - to serve up targeted advertisements to taxi passengers as they're ferried to their destination, based on their age and gender. The unsettling practice was discovered by Google privacy engineer Rosa Golijan, who posted a photo of a tablet she encountered when hopping into a taxi in Japan. Facial recognition technology is being deployed in airports, security cameras and in our phones. Now, Japan is using the tech to serve up targeted ads to passengers in taxis.

Dreams of dishing laundry duty to an in-house robot just got a little less hopeful after the company behind the automated assistant, Laundroid, filed for bankruptcy - effectively putting its bot to bed. According to Engadget, the company Seven Dreamers, which has worked to bring Laundroid to market since 2014, still owes 200 creditors about $20 million after its bankruptcy filing in Japan on April 23. In its Consumer Electronic Show (CES) debut in 2017, Seven Dreamers offered up what they purported would be an all-in-one laundry-folding and sorting machine that was able to take laundry, fold it, and in some cases organize it by color or owner and then deliver the final product to its wielder. The days of laundry folding robots got a little less hopeful after the makers of folding and sorting assistant Laundroid filed for bankruptcy. Users were meant to load the machine with clean dry clothes while a robot arm inside folds and sorts them.

When Seven Dreamers Laboratories Inc. unveiled its prototype laundry-folding robot in 2015, it generated a buzz, with people saying they couldn't wait to buy one if it ever went to market. But the AI-based tidying device dubbed Laundroid is apparently coming to an end before its commercial debut, as the Tokyo-based developer filed for bankruptcy Tuesday with the Tokyo District Court, citing insufficient funds to continue operations. A spokesperson for Seven Dreamers, a contest-winning startup that had received over ¥10 billion in funding, said development of robot is over for now. According to Teikoku Databank Ltd., a credit research company, Seven Dreamers Laboratories had accumulated ¥2.2 billion in debt as it struggled to ship the robot and invested heavily in research and development. After postponing its initial sales goal in fiscal 2017, it had to push back its goal for fiscal 2018, too.