Avi Mendelson
YOU?
Author Swipe
View article: Benchmarking Adversarial Patch Selection and Location
Benchmarking Adversarial Patch Selection and Location Open
Adversarial patch attacks threaten the reliability of modern vision models. We present PatchMap, the first spatially exhaustive benchmark of patch placement, built by evaluating over 1.5×108 forward passes on ImageNet validation images. Pa…
View article: Improving Resilience, Security, and Safety of Drones Through HTM-Based Adaptive Learning
Improving Resilience, Security, and Safety of Drones Through HTM-Based Adaptive Learning Open
This proposal suggests significantly improving the safety, the security, and the reliability of drones via a novel design technique that combines fault tolerance (FT) design methodology, the use of Hierarchical Operations that employ a dis…
View article: Robustness of Visual-Based Aerial Navigation to Real-World Adversarial Attacks
Robustness of Visual-Based Aerial Navigation to Real-World Adversarial Attacks Open
Imaging technologies are pivotal in the emerging aerial navigation ecosystem. However, these technologies are vulnerable to adversarial attacks. Current methods for enhancing the adversarial robustness of learned models are primarily based…
View article: Silent Tokens, Loud Effects: Padding in LLMs
Silent Tokens, Loud Effects: Padding in LLMs Open
Padding tokens are widely used in large language models (LLMs) to equalize sequence lengths during batched inference. While they should be fully masked, implementation errors can cause them to influence computation, and the extent of this …
View article: Efficient Context-Preserving Encoding and Decoding of Compositional Structures Using Sparse Binary Representations
Efficient Context-Preserving Encoding and Decoding of Compositional Structures Using Sparse Binary Representations Open
Despite their unprecedented success, artificial neural networks suffer extreme opacity and weakness in learning general knowledge from limited experience. Some argue that the key to overcoming those limitations in artificial neural network…
View article: Jailbreak Attack Initializations as Extractors of Compliance Directions
Jailbreak Attack Initializations as Extractors of Compliance Directions Open
Safety-aligned LLMs respond to prompts with either compliance or refusal, each corresponding to distinct directions in the model's activation space. Recent works show that initializing attacks via self-transfer from other prompts significa…
View article: The Use of Hierarchical Temporal Memory and Temporal Sequence Encoder for Online Anomaly Detection in Industrial Cyber-Physical Systems
The Use of Hierarchical Temporal Memory and Temporal Sequence Encoder for Online Anomaly Detection in Industrial Cyber-Physical Systems Open
This study introduces a novel, practical approach for designing a hierarchical online anomaly detection system for industrial cyber-physical systems. The proposed method utilizes the Hierarchical Temporal Memory (HTM) unsupervised learning…
View article: Sparse patches adversarial attacks via extrapolating point-wise information
Sparse patches adversarial attacks via extrapolating point-wise information Open
Sparse and patch adversarial attacks were previously shown to be applicable in realistic settings and are considered a security risk to autonomous systems. Sparse adversarial perturbations constitute a setting in which the adversarial pert…
View article: Hysteresis Activation Function for Efficient Inference
Hysteresis Activation Function for Efficient Inference Open
The widely used ReLU is favored for its hardware efficiency, {as the implementation at inference is a one bit sign case,} yet suffers from issues such as the ``dying ReLU'' problem, where during training, neurons fail to activate and const…
View article: AMED: Automatic Mixed-Precision Quantization for Edge Devices
AMED: Automatic Mixed-Precision Quantization for Edge Devices Open
Quantized neural networks are well known for reducing the latency, power consumption, and model size without significant harm to the performance. This makes them highly appropriate for systems with limited resources and low power capacity.…
View article: Electromigration-Aware Memory Hierarchy Architecture
Electromigration-Aware Memory Hierarchy Architecture Open
New mission-critical applications, such as autonomous vehicles and life-support systems, set a high bar for the reliability of modern microprocessors that operate in highly challenging conditions. However, while cutting-edge integrated cir…
View article: Teaching computing for complex problems in civil engineering and geosciences using big data and machine learning: synergizing four different computing paradigms and four different management domains
Teaching computing for complex problems in civil engineering and geosciences using big data and machine learning: synergizing four different computing paradigms and four different management domains Open
This article describes a teaching strategy that synergizes computing and management, aimed at the running of complex projects in industry and academia, in the areas of civil engineering, physics, geosciences, and a number of other related …
View article: Research in computing-intensive simulations for nature-oriented civil-engineering and related scientific fields, using machine learning and big data: an overview of open problems
Research in computing-intensive simulations for nature-oriented civil-engineering and related scientific fields, using machine learning and big data: an overview of open problems Open
This article presents a taxonomy and represents a repository of open problems in computing for numerically and logically intensive problems in a number of disciplines that have to synergize for the best performance of simulation-based feas…
View article: Electromigration-Aware Architecture for Modern Microprocessors
Electromigration-Aware Architecture for Modern Microprocessors Open
Reliability is a fundamental requirement in microprocessors that guarantees correct execution over their lifetimes. The reliability-related design rules depend on the process technology and device operating conditions. To meet reliability …
View article: Bimodal-Distributed Binarized Neural Networks
Bimodal-Distributed Binarized Neural Networks Open
Binary neural networks (BNNs) are an extremely promising method for reducing deep neural networks’ complexity and power consumption significantly. Binarization techniques, however, suffer from ineligible performance degradation compared to…
View article: AMED: Automatic Mixed-Precision Quantization for Edge Devices
AMED: Automatic Mixed-Precision Quantization for Edge Devices Open
Quantized neural networks are well known for reducing the latency, power consumption, and model size without significant harm to the performance. This makes them highly appropriate for systems with limited resources and low power capacity.…
View article: Bimodal Distributed Binarized Neural Networks
Bimodal Distributed Binarized Neural Networks Open
Binary Neural Networks (BNNs) are an extremely promising method to reduce deep neural networks' complexity and power consumption massively. Binarization techniques, however, suffer from ineligible performance degradation compared to their …
View article: Weisfeiler and Leman Go Infinite: Spectral and Combinatorial Pre-Colorings
Weisfeiler and Leman Go Infinite: Spectral and Combinatorial Pre-Colorings Open
Graph isomorphism testing is usually approached via the comparison of graph invariants. Two popular alternatives that offer a good trade-off between expressive power and computational efficiency are combinatorial (i.e., obtained via the We…
View article: Graph Representation Learning via Aggregation Enhancement
Graph Representation Learning via Aggregation Enhancement Open
Graph neural networks (GNNs) have become a powerful tool for processing graph-structured data but still face challenges in effectively aggregating and propagating information between layers, which limits their performance. We tackle this p…
View article: Contrast to Divide: Self-Supervised Pre-Training for Learning with Noisy Labels
Contrast to Divide: Self-Supervised Pre-Training for Learning with Noisy Labels Open
The success of learning with noisy labels (LNL) methods relies heavily on the\nsuccess of a warm-up stage where standard supervised training is performed\nusing the full (noisy) training set. In this paper, we identify a "warm-up\nobstacle…
View article: NICE: Noise Injection and Clamping Estimation for Neural Network Quantization
NICE: Noise Injection and Clamping Estimation for Neural Network Quantization Open
Convolutional Neural Networks (CNNs) are very popular in many fields including computer vision, speech recognition, natural language processing, etc. Though deep learning leads to groundbreaking performance in those domains, the networks u…
View article: NICE: Noise Injection and Clamping Estimation for Neural Network Quantization
NICE: Noise Injection and Clamping Estimation for Neural Network Quantization Open
Convolutional Neural Networks (CNNs) are very popular in many fields including computer vision, speech recognition, natural language processing, etc. Though deep learning leads to groundbreaking performance in those domains, the networks u…
View article: Contrast to Divide: Self-Supervised Pre-Training for Learning with Noisy Labels
Contrast to Divide: Self-Supervised Pre-Training for Learning with Noisy Labels Open
The success of learning with noisy labels (LNL) methods relies heavily on the success of a warm-up stage where standard supervised training is performed using the full (noisy) training set. In this paper, we identify a "warm-up obstacle": …
View article: Asymmetric aging effect on modern microprocessors
Asymmetric aging effect on modern microprocessors Open
Reliability is a crucial requirement in any modern microprocessor to assure correct execution over its lifetime. As mission critical components are becoming common in commodity systems; e.g., control of autonomous cars, the demand for reli…
View article: Early-Stage Neural Network Hardware Performance Analysis
Early-Stage Neural Network Hardware Performance Analysis Open
The demand for running NNs in embedded environments has increased significantly in recent years due to the significant success of convolutional neural network (CNN) approaches in various tasks, including image recognition and generation. T…
View article: Towards Designing a Secure RISC-V System-on-Chip: ITUS
Towards Designing a Secure RISC-V System-on-Chip: ITUS Open
A rising tide of exploits, in the recent years, following a steady discovery of the many vulnerabilities pervasive in modern computing systems has led to a growing number of studies in designing systems-on-chip (SoCs) with security as a fi…
View article: FlexWatts: A Power- and Workload-Aware Hybrid Power Delivery Network for Energy-Efficient Microprocessors
FlexWatts: A Power- and Workload-Aware Hybrid Power Delivery Network for Energy-Efficient Microprocessors Open
Modern client processors typically use one of three commonly-used power delivery network (PDN): 1) motherboard voltage regulators (MBVR), 2) integrated voltage regulators (IVR), and 3) low dropout voltage regulators (LDO). We observe that …
View article: Self-Supervised Learning for Large-Scale Unsupervised Image Clustering
Self-Supervised Learning for Large-Scale Unsupervised Image Clustering Open
Unsupervised learning has always been appealing to machine learning researchers and practitioners, allowing them to avoid an expensive and complicated process of labeling the data. However, unsupervised learning of complex data is challeng…
View article: Feature Map Transform Coding for Energy-Efficient CNN Inference
Feature Map Transform Coding for Energy-Efficient CNN Inference Open
Convolutional neural networks (CNNs) achieve state-of-the-art accuracy in a variety of tasks in computer vision and beyond. One of the major obstacles hindering the ubiquitous use of CNNs for inference on low-power edge devices is their hi…