After launching it in Go stores and then bringing it to larger Fresh supermarkets, Amazon's cashierless "Just Walk Out" tech will soon arrive in two Whole Foods locations. The service, which lets you pick up goods from shelves and (yep) just walk out, is coming to new stores in Washington DC and Sherman Oaks, California next year, the company announced. "By collaborating with Amazon to introduce Just Walk Out shopping at these two Whole Foods Market stores, our customers will be able to... save time by skipping the checkout line," said Whole Foods co-founder John Mackey. As we've detailed previously, Just Walk Out uses computer vision, sensors and AI to let you walk into a store, sign in with an app, fill up your bags and leave without the need to join a checkout line. On top of using the tech in its own Go and Fresh stores, Amazon signed a deal last year to license its technology to third-party retailers.

Long Short-Term Memory (LSTM) recurrent networks are frequently used for tasks involving time-sequential data such as speech recognition. However, it is difficult to deploy these networks on hardware to achieve high throughput and low latency because the fully connected structure makes LSTM networks a memory-bounded algorithm. Previous LSTM accelerators either exploited weight spatial sparsity or temporal activation sparsity. This paper proposes a new accelerator called "Spartus" that exploits spatio-temporal sparsity to achieve ultra-low latency inference. The spatial sparsity is induced using our proposed pruning method called Column-Balanced Targeted Dropout (CBTD), which structures sparse weight matrices for balanced workload. It achieved up to 96% weight sparsity with negligible accuracy difference for an LSTM network trained on a TIMIT phone recognition task. To induce temporal sparsity in LSTM, we create the DeltaLSTM by extending the previous DeltaGRU method to the LSTM network. This combined sparsity simultaneously saves on the weight memory access and associated arithmetic operations. Spartus was implemented on a Xilinx Zynq-7100 FPGA. The Spartus per-sample latency for a single DeltaLSTM layer of 1024 neurons averages 1 us. Spartus achieved 9.4 TOp/s effective batch-1 throughput and 1.1 TOp/J energy efficiency, which, respectively, are 4X and 7X higher than the previous state-of-the-art.

Krasner is the author of Department of Neurology at the University the "CSCW '86 Summary Report." of California at Davis.

Even though sues raised during the workshop sponsored emerged during the workshop. In many situations, users are not sufficiently defined or Seattle, Washington. Issues include the need guidance and counseling in order understood to be amenable to traditional definition of coupled systems, motivations to solve the problem at hand. In control system--one that combines such situations, users often need help techniques from artificial intelligence in determining which specific algorithm (AI), control theory, and operations or technique should be research (Kowalik et al. 1986). In other situations, traditional techniques to perform the need is more basic--for guidance in many routine tasks, sophisticated determining whether the problem at hand can be solved and, if so, whether techniques are needed to handle many the resources that can be brought to of the humanlike functions.