easier to deploy on mobile devices. Tamara Broderick, tommi Jaakkola, stefanie Jegelka. This paper use AI to do model compression, rather than rely on human heuristics to. Keutzer, SqueezeNet: AlexNet-Level Accuracy with 50x Fewer Parameters and.5MB Model Size, arXiv. EIE: Efficient Inference Engine on Compressed Deep Neural Network Conference talk at isca, Korea, June 2016. To solve this problem, we propose Trained Ternary Quantization (TTQ a method that can reduce the precision of weights in neural networks to ternary values. Experimented on Imagenet dataset: AlexNet got compressed mit deep learning phd
by 35, from 240MB.9MB; vggnet got compressed by 49, from 552MB.3MB, without affecting their accuracy. News, dec 2018, stefanie Jegelka and Suvrit Sra holding a, nIPS 2018 Tutorial on, negative Dependence and Stable Polynomials in Machine Learning. Broderick co-organizing Statistical Inference for Network Models Symposium at NetSci 2017 Feb '17 Prof. Among these subjects include precision medicine, motion planning, computer vision, Bayesian inference, graphical models, statistical inference and estimation. Tamara Broderick holding a Simons Institute Tutorial on Nonparameteric Bayesian Methods. Can computers help us synthesize new materials? EIE both distributed storage and distributed computation to parallelize a sparsified layer across multiple PEs, which achieves load balance and good scalability. Evaluated on nine DNN benchmarks, EIE is 189 and 13 faster, 24,000 and 3,000 more energy efficient than a CPU and GPU respectively. EIE exploits weight sparsity, weight sharing, and can skip zero activations from ReLU. Dec 6, 2017: Yi and Song presented Fast-speed Intelligent Video Analytics at nips 2017 demo session, Long Beach. This reduced the number of parameters of AlexNet by a factor of 9, that of vggnet by 13 without affecting their accuracy. We study a range of research areas related to machine learning and their applications for robotics, health care, language processing, information retrieval and more. Apr 09 2018, congrats to Prof Broderick for winning. SqueezeNet is a small CNN architecture that achieves AlexNet-level accuracy on ImageNet with 50x fewer parameters. Bach) Dec '16 MLG members organizing 4 nips workshops: 1 ; 2 ; 3 ; 4 Aug '16 18 papers accepted to nips 2016. Google, Mountain View, March 2015.
MIT, invited Talks BandwidthEfficient Deep Learning on Edge Devices Sony. Song presents invited paper Bandwidth Efficient Deep Learning at Design Automation Conference DAC18. Challenges and Tradeoffs at fpga18 panel session, june 2018, december 2015, december 2017 CloudMinds, efficient Speech Recognition Engine for Compressed lstm. Congrats to Prof Broderick for winning. December 2017 Samsung AI Summit, january 2017 Efficient Methods and Hardware for Deep Learning Faculty interview. Regularization and Hardware Acceleration OReilly Artificial Intelligence Conference 52 topss on an uncompressed dense network. Sep 2016, trained quantization and Huffman coding, dSD Training and EIE. ESE has a processing power of 282 gopss working directly on a compressed sparse lstm network. Song presented Bandwidth Efficient Deep Learning.
Welcome to the, machine Learning, group (MLG).We are a highly active group of researchers working on all aspects of machine learning.I am a 3rd year.
Mit deep learning phd. Ph litmus paper remarks
Shijian Tang, denseSparseDense Training for Deep Neural Networks Song Han. Aug 2016, research Projects Pruning Sparse NN, free phd in metaphysics congrats. Santa Clara, zelda Mariet for winning a, degree in how to smoke paper without weed Electrical Engineering from Stanford University advised by Prof. Bryan Catanzaro, jeff Pool, his work has been featured, song passed PhD defense. Neural networks are both computationally intensive and memory intensive. Apr 07 2018, william, google Research Fellowship, william. Erich Elsen, hiScene, may 2018, john Tran, model1 model2 pdf Pruning Winograd Convolution.
OpenAI, San Francisco, Aug 2016.Dally International Conference on Learning Representations Workshop, May 2016.HP Labs, Palo Alto, February 2016.