Новости ИСП РАН

Новости ИСП РАН

30 Марта, 2018

С 2 по 4 апреля в ИСП РАН будут проходить лекции сотрудника университета Технион

С 2 по 4 апреля в ИСП РАН будут проходить лекции аспиранта университета Технион (Израильский технологический институт) Chaim Baskin (Хаим Баскин).

Расписание лекций:

02 апреля 2018 г.

15:00-17:00 Intro to deep learning (Algorithms and hardware aspects).

03 апреля 2018 г.

11:00-13:00 Convolutional Neural Networks for classification (History+ state of the art architectures).

15:00-16:00 Auto encoders and other regression architectures.

04 апреля 2018 г.

11:00-13:00 Into to Generative models.

15:00-15:45 Семинар по теме "Recurrent Neural Networks for Computer Vision and Neural Language Programming (NLP)".


Deep neural networks (DNNs) are used by different applications that are executed on a range of computer architectures, from IoT devices to supercomputers. The footprint of these networks is huge as well as their computational and communication needs. In order to ease the pressure on resources, research indicates that in many cases a low precision representation (1-2 bit per parameter) of weights and other parameters can achieve similar accuracy while requiring less resources. Using quantized values enables the use of FPGAs to run NNs, since FPGAs are well fitted to these primitives; e.g., FPGAs provide efficient support for bitwise operations and can work with arbitrary-precision representation of numbers. This paper presents a new streaming architecture for running QNNs on FPGAs. The proposed architecture scales out better than alternatives, allowing us to take advantage of systems with multiple FPGAs. We also included support for skip connections, that are used in state-of-the art NNs, and shown that our architecture allows to add those connections almost for free. All this allowed us to implement an 18-layer ResNet for $224\times224$ images classification, achieving $57.5\%$ top-1 accuracy. In addition, we implemented a full-sized quantized AlexNet. In contrast to previous works, we use 2-bit activations instead of 1-bit ones, which improves AlexNet's top-1 accuracy from $41.8\%$ to $51.03\%$ for the ImageNet classification. Both AlexNet and ResNet can handle 1000-class real-time classification on an FPGA.

Our implementation of ResNet-18 consumes $5\times$ less power and is $4\times$ slower for ImageNet, when compared to the same NN on the latest Nvidia GPUs. Smaller NNs, that fit a single FPGA, are running faster then on GPUs on small ($32\times32$) inputs, while consuming up to $20\times$ less energy and power.

The seminar talk will cover paper "Streaming Architecture for Large-Scale Quantized Neural Networks on an FPGA-Based Dataflow Platform" that have been accepted to RAW 2018.

Все новости