A5088051631- Advanced Neural Network Applications
- Neural Networks and Applications
- Adversarial Robustness in Machine Learning
- Physical Unclonable Functions (PUFs) and Hardware Security
- Cryptographic Implementations and Security
- Advancements in PLL and VCO Technologies
- Anomaly Detection Techniques and Applications
- Integrated Circuits and Semiconductor Failure Analysis
- Advancements in Semiconductor Devices and Circuit Design
- Scientific Computing and Data Management
- Ferroelectric and Negative Capacitance Devices
- Photonic and Optical Devices
- Security and Verification in Computing
- Fault Detection and Control Systems
- Smoking Behavior and Cessation
- Software System Performance and Reliability
- Advanced DC-DC Converters
- COVID-19 Digital Contact Tracing
- Low-power high-performance VLSI design
- Photovoltaic System Optimization Techniques
- COVID-19 epidemiological studies
- Semiconductor materials and devices
- COVID-19 Pandemic Impacts
- Machine Learning and Data Classification
- Global Cancer Incidence and Screening
University of California, San Diego
2021-2025
Fermi National Accelerator Laboratory
2025
UC San Diego Health System
2024
Washington University in St. Louis
2023
New York City Department of Health and Mental Hygiene
2021
Universiti Putra Malaysia
2005
Abstract Background Although colorectal cancer screening has contributed to decreased incidence and mortality, disparities are present by race/ethnicity. The Citywide Colon Cancer Control Coalition (C5) NYC Department of Health Mental Hygiene (DOHMH) promoted colonoscopy from 2003 on, hypothesized future reductions in CRC incidence, mortality racial/ethnic disparities. Methods We assessed annual percent change (APC) stage rates through 2016 a longitudinal cross-sectional study NY State...
In this paper, we propose a method to perform empirical analysis of the loss landscape machine learning (ML) models. The is applied two ML models for scientific sensing, which necessitates quantization be deployed and are subject noise perturbations due experimental conditions. Our allows assessing robustness such effects as function precision under different regularization techniques -- crucial concerns that remained underexplored so far. By investigating interplay between performance,...
State tobacco quitlines are delivering cessation assistance through an increasingly diverse range of channels. However, offerings vary from state to state, many smokers unaware what is available, and it not yet clear how much demand exists for different types assistance. In particular, the online digital interventions among low-income smokers, who bear a disproportionate burden tobacco-related disease, well understood.
Voltage fluctuation sensors measure minute changes in an FPGA power distribution network, allowing attackers to extract information from concurrently executing computations. Previous voltage make assumptions about the co-tenant computation and require attacker have a priori access or system knowledge tune sensor parameters statically. Additionally, prior use of proprietary vendor intellectual property do not provide guidance on migration other vendors. We present open-source design Tunable...
Voltage fluctuation sensors measure minute changes in an FPGA power distribution network, allowing attackers to extract information from concurrently executing computations. Previous voltage make assumptions about the co-tenant computation and require attacker have a priori access or system knowledge tune sensor parameters statically. We present open-source design of Tunable Dual-Polarity Time-to-Digital Converter, which introduces three dynamically tunable that optimize signal measurement,...
Studies examining individual-level changes in protective behaviors over time association with community-level infection and self or close-contact SARS-CoV-2 are limited. We analyzed overall demographic specific week-to-week COVID-19 their infections (regional case counts close contacts). Data were collected through 37 consecutive weekly surveys from 10/17/2021 - 6/26/2022. Our survey panel included 212 individuals living working St. Louis City County, Missouri, U.S.A. Frequency of...
With more scientific fields relying on neural networks (NNs) to process data incoming at extreme throughputs and latencies, it is crucial develop NNs with all their parameters stored on-chip. In many of these applications, there not enough time go off-chip retrieve weights. Even so, memory such as DRAM does have the bandwidth required fast being produced (e.g., every 25 ns). As such, latency requirements architectural implications for hardware intended run NNs: 1) NN must fit on-chip, 2)...
Remote attackers can recover "FPGA pentimento" - long-removed data belonging to a prior user or proprietary design image on cloud FPGA. Just as pentimento of painting be exposed by infrared imaging, FPGA pentimentos signal timing sensors. The constituting an is imprinted the device through bias temperature instability effects underlying transistors. Measuring this degradation using time-to-digital converter allows attacker (1) extract details keys from encrypted available AWS marketplace and...
Edge computation often requires robustness to faults, e.g., reduce the effects of transient errors and function correctly in high radiation environments. In these cases, edge device must be designed with fault tolerance as a primary objective. FKeras is tool that helps design fault-tolerant neural networks (NNs) run entirely on chip meet strict latency resource requirements. provides metrics give bit-level ranking NN weights respect their sensitivity faults. includes guide efficient...
Extreme data rate scientific experiments create massive amounts of that require efficient ML edge processing. This leads to unique validation challenges for VLSI implementations algorithms: enabling bit-accurate functional simulations performance in experimental software frameworks, verifying those models are robust under extreme quantization and pruning, ultra-fine-grained model inspection fault tolerance. We discuss approaches developing validating reliable algorithms at the such strict...
Deep neural networks use skip connections to improve training convergence. However, these are costly in hardware, requiring extra buffers and increasing on- off-chip memory utilization bandwidth requirements. In this article, we show that can be optimized for hardware when tackled with a hardware-software codesign approach. We argue while network’s needed the network learn, they later removed or shortened provide more hardware-efficient implementation minimal no accuracy loss. introduce...
We present our development experience and recent results for the MLPerf Tiny Inference Benchmark on field-programmable gate array (FPGA) platforms. use open-source hls4ml FINN workflows, which aim to democratize AI-hardware codesign of optimized neural networks FPGAs. design implementation process keyword spotting, anomaly detection, image classification benchmark tasks. The resulting hardware implementations are quantized, configurable, spatial dataflow architectures tailored speed...
Extreme data rate scientific experiments create massive amounts of that require efficient ML edge processing. This leads to unique validation challenges for VLSI implementations algorithms: enabling bit-accurate functional simulations performance in experimental software frameworks, verifying those models are robust under extreme quantization and pruning, ultra-fine-grained model inspection fault tolerance. We discuss approaches developing validating reliable algorithms at the such strict...
Deep neural networks use skip connections to improve training convergence. However, these are costly in hardware, requiring extra buffers and increasing on- off-chip memory utilization bandwidth requirements. In this paper, we show that can be optimized for hardware when tackled with a hardware-software codesign approach. We argue while network's needed the network learn, they later removed or shortened provide more efficient implementation minimal no accuracy loss. introduce Tailor, tool...
Background The U.S. Surgeon General and others have emphasized a critical need to address COVID-19 misinformation protect public health. In St. Louis, MO, we created iHeard STL , community-level surveillance response system. This paper reports methods findings from its first year of operation. Methods We assembled panel over 200 community members who answered brief, weekly mobile phone surveys share information they heard in the last seven days. Based on their responses, prioritized threats....
Uninterrupted power supply (UPS) systems are used as one solution of quality problems and to provide ultimate protection for disturbances such blackouts brownouts. Many UPS suffer from poor output voltage regulation especially with heavy loads. This work is aimed design implement the hardware system capable producing continuous constant 230 V/sub ac/, 50 Hz supply. A feedback loop has been implemented using microcontroller adjust dc level supplying inverter. At end implementation, tests have...
In this community review report, we discuss applications and techniques for fast machine learning (ML) in science -- the concept of integrating power ML methods into real-time experimental data processing loop to accelerate scientific discovery. The material report builds on two workshops held by Fast Science covers three main areas: across a number domains; training implementing performant resource-efficient algorithms; computing architectures, platforms, technologies deploying these...
Deep neural networks employ skip connections – identity functions that combine the outputs of different layers-to improve training convergence; however, these are costly to implement in hardware. In particular, for inference accelerators on resource-limited platforms, they require extra buffers, increasing not only on- and off-chip memory utilization but also bandwidth requirements. Thus, a network has costs more deploy hardware than one none. We argue that, certain classification tasks,...
Cloud FPGAs strike an alluring balance between computational efficiency, energy and cost. It is the flexibility of FPGA architecture that enables these benefits, but very same exposes new security vulnerabilities. We show a remote attacker can recover "FPGA pentimenti" - long-removed secret data belonging to prior user cloud FPGA. The sensitive constituting pentimento analog imprint from bias temperature instability (BTI) effects on underlying transistors. demonstrate how this slight...
We present FKeras, an open-source tool that uses Hessian information to quickly find which parameters in a neural network are sensitive radiation faults, reducing the usual 200% resource overhead needed protect them.
Residual networks (ResNets) employ skip connections in their -- reusing activations from previous layers to improve training convergence, but these create challenges for hardware implementations of ResNets. The must either wait be processed before processing more incoming data or buffer them elsewhere. Without connections, ResNets would hardware-efficient. Thus, we present the teacher-student learning method gradually prune away all a ResNet's constructing network call NonResNet. We show...
Side-channel leakage poses a major security threat in multi-tenant FPGA environments. One tenant can instantiate voltage fluctuation sensor that measures minute changes the power distribution network (PDN) and infer information about co-tenant computation data. This work presents Tunable Dual-Edged Time-to-Digital Converter (TDC) - with two unique elements: first, it has ability to tune sample duration, phase, frequency more effectively extract co-located computation; second, captures both...