In this recent work, we show that neuromorphic systems based on memristor crossbars can easily be attacked (adversarial evasion attack) using power analysis, even if the attacker has no knowledge about the dataset used to train the neural network. The number of power measurements needed can be made relatively small through Bayesian optimization. https://lnkd.in/gj3F_wDW
brainlab
研究服务
Rochester，NY 303 位关注者
We are a multidisciplinary research lab at RIT focused on defining the next generation of ultra-efficient AI systems.
关于我们
We are a multidisciplinary research lab at Rochester Institute of Technology focused on defining the next generation of ultra-efficient AI systems.
- 网站
-
https://www.rit.edu/brainlab
brainlab的外部链接
- 所属行业
- 研究服务
- 规模
- 11-50 人
- 总部
- Rochester，NY
- 类型
- 教育机构
地点
-
主要
获取路线
US，NY，Rochester，14623
brainlab员工
动态
-
It's an absolute pleasure to be a part of this amazing and diverse community of researchers. We are thrilled that we were able to contribute to this authoritative work that helps build consensus on the state of neuromorphic computing and how to scale them for large-scale AI applications.
Published in #Nature today. Over the course of one year, a team of 23 researchers from 17 organizations across the world, worked diligently thinking about the state of neuromorphic computing and how to achieve largescale systems. We couldn’t be more excited to share it with the world!? When we began charting this course, we questioned whether we could approach the topic impartially and apply a critical lens to our field. What really made this possible was the close-knit teamwork across researchers from academia, industry, and national labs. In a field as interdisciplinary as neuromorphic computing, where no single solution fits all, this kind of cross-disciplinary collaboration is what leads to real, meaningful progress. Collaborative Team: Craig vineyard, Catherine Schuman, Tej Pandit, Joe Hays, Cory Merkel, Suma George Cardwell, Brad Aimone, Chiara Bartolozzi, Cliff Young, Garrick Orchard, Rajkumar Kubendran, Ryad Benosman, Chetan Singh Thakur, Gert Cauwenberghs, Melika Payvand,Christian Mayr, Hector A. Gonzalez, Sonia Buckley, Amitava Majumdar, Anand Subramoney, Shruti R. Kulkarni, Steve Furber If you are curious to learn more, full paper link here: https://lnkd.in/gjf6ADYh We are open to learning of any blindspots in the course. Credit: Ideas originated during and after the #NSF Largescale Neuromorphic Computing workshop in 2022, held in Knoxville. Roots for The Neuromorphic Commons- THOR started here.
-
-
We have a couple new papers you may find interesting! 1.) "Activation Function Perturbations in Artificial Neural Networks Effects on Robustness" - Led by Brain Lab member and Math Modeling PhD student Justin S., this paper is one of the first to explore the impact of activation function perturbations on adversarial robustness: https://lnkd.in/gUDaPQgk 2.) "Realizing Linear Synaptic Plasticity in Electric Double Layer-Gated Transistors for Improved Predictive Accuracy and Efficiency in Neuromorphic Computing" - Led by Nithil Harris Manimaran and Ke Xu's team, this paper applies novel synaptic devices to energy-efficient neuromorphic computing. https://lnkd.in/gY7w6beJ
-
Very exciting news from friend of brainlab Alexander Ororbia II! https://lnkd.in/epkpzK28
RIT professor proposes new way to make artificial intelligence smarter and greener
rit.edu
-
We are excited to announce some new preliminary work on implementing contrastive learning using Hebbian rules in memristor-based neuromorphic systems. The work is in collaboration with Alexander Ororbia II and will be presented at the IEEE Workshop on Signal Processing Systems. Check out the pre-print here: https://lnkd.in/gZdtWkFS
Contrastive Learning in Memristor-based Neuromorphic Systems
arxiv.org
-
Congratulations to RIT Electrical and Computer Engineering Ph.D. student and brainlab collaborator Katsuaki Nakano on his paper, "Trustworthy and Robust Machine Learning for Multimedia: Challenges and Perspectives," to be presented at IEEE Multimedia Information Processing and Retrieval 2024 (MIPR'24). The work is also in collaboration with Michael Zuzak and Alexander C. Loui in RIT Computer Engineering.
-
Congratulations to RIT EME student Karsten Bergthold and ECE student Hagar Hossam on their work "Throughput Optimization for Time-Domain Neuromorphic Computing," accepted for presentation in MWSCAS 2024! The work, in collaboration with RIT EME's Tejasvi Das, explores techniques like wave pipelining to push data through memristor-based neuromorphic systems faster.
-
Congratulations to Alex Henderson and team for the new paper "Memristor Based Liquid State Machine with Method for In-Situ Training," in IEEE TNANO. We were happy to be able to play a role contributing to this cool work! https://lnkd.in/gsq8Hdbu
Memristor Based Liquid State Machine With Method for In-Situ Training
ieeexplore.ieee.org