|
|
|
I am a PhD candidate in the Department of Computer Science and Engineering (CSE) at the University of California, San Diego (UCSD). I received my M.S. and B.Sc degrees from the School of Electrical and Computer Engineering (ECE) at the University of Tehran (UT), Tehran in 2014 and 2011 respectively. Currently, I am a recipient of the Jacobs School Powell Fellowship at UC San Diego.
I am a member of System Energy Efficiency Lab (SeeLab) working with Professor Tajana Rosing. My research focuses on computer architecture, machine learning and brain-inspired computing. I am working on the technological, design and algorithmic aspects of these project collaborating with UC Berkeley's Wireless Research Center and UCSD ACES Lab. You can find my CV here.
|
Projects
Approximate Computing:
We propose a novel architecture with software and hardware support for approximate computing. Hardware components are enhanced with ability to dynamically adapt approximation at a quantifiable and controllable cost in terms of accuracy. Software services complement hardware to ensure user’s perception is not compromised, while maximizing the energy savings due to approximations. The changes to hardware design include approximation enabled CPU, GPU, accelerators (DSPs) and storage.
We propose a novel architecture with software and hardware support for approximate computing. Hardware components are enhanced with ability to dynamically adapt approximation at a quantifiable and controllable cost in terms of accuracy. Software services complement hardware to ensure user’s perception is not compromised, while maximizing the energy savings due to approximations. The changes to hardware design include approximation enabled CPU, GPU, accelerators (DSPs) and storage.
Machine Learning Accelerator:
Many machine learning and cognitive applications process large data set collected in the Internet of Things domain. Running the data-intensive workloads on traditional cores results in high energy consumption and slow processing speed. This inefficiency stems from the large amount of data movements between memory and processing units in the traditional architectures. In addition, many cognitive systems, e.g., big data analysis and healthcare applications, assimilate the data using deep neural networks and classification algorithms which are also computationally expensive in the traditional Von Neumann architecture. We design novel memory architectures which completely replaces the traditional processing units and performs brain-inspired computations such as deep neural networks and hyperdimensional computing.
Many machine learning and cognitive applications process large data set collected in the Internet of Things domain. Running the data-intensive workloads on traditional cores results in high energy consumption and slow processing speed. This inefficiency stems from the large amount of data movements between memory and processing units in the traditional architectures. In addition, many cognitive systems, e.g., big data analysis and healthcare applications, assimilate the data using deep neural networks and classification algorithms which are also computationally expensive in the traditional Von Neumann architecture. We design novel memory architectures which completely replaces the traditional processing units and performs brain-inspired computations such as deep neural networks and hyperdimensional computing.
Brain-Inspired Hyperdimensional Computing:
The mathematical properties of high-dimensional spaces show remarkable agreement with behaviors controlled by the brain. Brain-inspired hyperdimensional (HD) computing explores the emulation of cognition by computing with hypervectors as an alternative to computing with numbers. HD computing provides an opportunity for robust, general and scalable model of computing using hypervectors in thousand dimensions. We have developed the application of hyperdimensional computing on text analysis, voice analysis, biosignal processing and image classification, while achieving high levels of classification accuracy, energy efficiency, and robustness. In all these designs HD can provide comparable or better accuracy than existing machine learning algorithm to solve the same task.
The mathematical properties of high-dimensional spaces show remarkable agreement with behaviors controlled by the brain. Brain-inspired hyperdimensional (HD) computing explores the emulation of cognition by computing with hypervectors as an alternative to computing with numbers. HD computing provides an opportunity for robust, general and scalable model of computing using hypervectors in thousand dimensions. We have developed the application of hyperdimensional computing on text analysis, voice analysis, biosignal processing and image classification, while achieving high levels of classification accuracy, energy efficiency, and robustness. In all these designs HD can provide comparable or better accuracy than existing machine learning algorithm to solve the same task.
Emerging Memory/Logic Design:
Designing a memory subsystem is challenging because it involves balancing among multiple goals. The goals include fast access time, programmability, high bandwidth, low power, low cost, reliability, and security. Some of these goals may contradict one another, thus it is important to strike a good balance given the target market of a system. We exploit the feature of emerging technologies, e.g. emerging transistors and non-volatile memories, to design an ultra-efficient and stable memories for embedded applications in circuit, architecture and application levels.
Designing a memory subsystem is challenging because it involves balancing among multiple goals. The goals include fast access time, programmability, high bandwidth, low power, low cost, reliability, and security. Some of these goals may contradict one another, thus it is important to strike a good balance given the target market of a system. We exploit the feature of emerging technologies, e.g. emerging transistors and non-volatile memories, to design an ultra-efficient and stable memories for embedded applications in circuit, architecture and application levels.
Mentorship:
Based on my research and students interest, we opened a new sub-group in SeeLab, Alternative Computing and Memory Systems (ACMS), where I am closely mentoring several SMART graduate and undergrad students:
|
Current Members:
- Joonseop Sim, CSE PhD, Project: Emerging Memory/Circuit Design - Daniel Perioni, CSE PhD, Project: Approximate computing - Saransh Gupta, ECE MS, Project: In-memory computing - Harveen Kaur, CSE MS, Project: Deep learning - Sahil Sharma, CSE MS, Project: Map reduce and query processing - Max Masich, CSE MS, Project: Neural Network - Pushen Wang, CSE MS, Project: Framework to accelerate supervised learning - Minxuan Zhou, CSE MS, Project: Graph processing - Chihheng Tzang, CSE MS, Project: Approximate logic - Deqian Kong, CSE BSc, Project: Brain-Inspired computing - Thomas Worley, CSE BSc, Project: Classification and clustering acceleration - Tarek Nassar, ECE BSc, Project: GPU-based deep neural network acceleration - Seven Wu, CSE BSc, Project: Text classification - Shuo Li, CSE BSc, Project: Image-like text classification - Jiaxiao Zhou, CSE BSc, Project: Self learning in high-dimensional space Alumni Members: - Atl Arredondo (BSc CSE), Data science engineer at slack Inc., Jan 2016 - Debanjan Chatterjee (MS ECE), Engineer at Citadel Information Services Inc., Jan 2016 - Yan Cheng (BSc CSE), Graduate student at CSE, UCSD., July 2016 - John F Hwang (BSc CSE), Engineer at Parsons Corporation, Sept 2016 - Zhipeng Yan, CSE MS, UCSD - Moyuan Huang, CSE MS, UCSD |
|
What's New?
New! June 2017. Accepted paper in ICCAD! New! June 2017. Paper presentation at DAC'17! New! June 2017. Accepted paper in D&T. Thanks Dr. Abbas Rahimi and prof. J. M. Rabaery! New! June 2017. My mentee Max Masich defensed his master thesis! New! June 2017. My mentee Deqian Kong selected as best undergrad researcher at CSE [Picture]! May 2017. Accepted papers in Techcon conference! May 2017. Program committee of IEEE SOC'17 and IEEE PRIM'17! May 2017. accepted paper in ISLPED! March 2017. ISQED paper presentation, Santa Clara! March 2017. NVMW paper presentation, La Jolla! Feb 2017. Two accepted papers in DAC 2017! Jan 2017. Accepted paper in EEE Transactions on Multi-Scale Computing Systems. Dec 2016. Accepted CAP Processor paper in ISQED 2017. Nov 2016. Accepted CPU approximation papers in IEEE TETC. Nov 2016. Accepted papers in DATE 2017! Thanks prof. F. Koushanfar group at ECE. Oct 2016. Accepted paper in HPCA 2017! Thanks our collaborator at UC Berkeley (Dr. Abbas Rahimi and prof. J. M. Rabaery) Sept 2016. Accepted paper in ASPDAC 2017! Aug 2016. ICCD'16 paper selected as top ranked paper for publishing in IEEE TETC! Aug 2016. Accepted paper in ICCD 2016! June 2016. DAC paper presentation, Austin! June 2016. Being nominated for the Outstanding Graduate Student Award at UCSD! June 2016. Accepted paper in NVMSA 2016! June 2016. Accepted paper in Techcon conference 2016! May 2016. GLSVLSI paper presentation, Boston! May 2016. Accepted paper in MEMSYS Conference! Washington DC May 2016. Defensed my proposal!! (PhD Candidate) May 2016. Accepted paper in ISLPED for long presentation, San Francisco! April 2016. Accepted paper in IEEE Transaction on Emerging Topics Computing! April 2016. Research Expo'16 poster presentation. March 2016. ISQED paper presentation, Santa Clara! February 2016. Accepted GLSVLSI 2016 papers! January 2016. "From non-volatile memory to approximate computing" talk at UCSD seminar class [Slide]. December 2015. Accepted ISQED 2016 papers! November 2015. Accepted DATE 2016 papers! November 2015. ICCAD paper presentation [Slide], Austin! July 2015. Nanoarch conference paper presentation, Boston! June 2015. Accepted ICCAD 2015 paper!! June 2015. DAC conference paper presentation, San Francisco! May 2015. Research exam passed! April 2015. Research Expo'15 poster presentation. March 2015. NVMW poster presentation! |
Collaborators and Sponsors:
|
