arXiv subjects

Richard George

Publications and source records attributed to Richard George.

9 recordsLinked to original sources

arXiv2026

A Resource-Efficient CNN-Based EEG Auditory Attention Decoding ASIC

Following a target speaker in a noisy environment, commonly known as the cocktail party problem, remains particularly challenging for cochlear implant (CI) users. Recent studies have explored EEG-based auditory attention decoding (AAD) using neural networks to enhance hearing assistance. This paper presents a resource-efficient ASIC for real-time EEG-based auditory attention decoding by integrating a quantized CNN inference engine and a Pearson-correlation classifier. The proposed architecture employs streaming execution, on-chip buffering, and memory-efficient dataflow to reduce hardware cost while maintaining real-time performance. The proposed ASIC has been fully implemented in GF22FDX 22-nm CMOS technology, occupying a total silicon area of 2.09 mm$^2$(1264$\mu$m x 1654$\mu$m), with the CNN inference engine and streaming classification engine requiring only 0.076 mm$^2$. Operating at a core voltage of 0.55 V, the design achieves a power consumption of 0.4941 mW and an inference latency of 7.34 ms, providing an energy-efficient hardware platform for EEG-based auditory attention decoding in hearing-assistance applications.

cs.AR↗

arXiv2026

Event-based Neural Decoding for Neuroprosthetic Motor Control

A substantial number of patients experience diminished mobility due to disabilities, diseases, or accidents. Although modern prostheses, powered by deep neural networks, hold the promise of significantly enhancing the quality of life for these individuals, their widespread adoption is hindered by significant latency, energy consumption, and spatial requirements. Wired connections to external high-performance processors restrict patient mobility, while wireless connections limit the volume of information that can be transmitted to these processors. Spiking neural networks offer the potential for compressed communication and low-power inference, yet they often lag behind state-of-the-art deep learning models in various applications. In this study, we propose a high-performance neural decoding method that effectively balances task performance and efficiency. An eventbased gated recurrent unit generates a sparse communication pattern with graded spikes, surpassing classical spiking neural networks in terms of task performance. Utilising an efficient training method and sparse inference, our model presents new opportunities for on-device neural decoding.

cs.LG↗

arXiv2024

Performance of iDQ ahead of LIGO, Virgo, and KAGRA's fourth observing run

The gravitational wave detectors used by the LIGO Scientific Collaboration, and the Virgo Collaboration are incredibly sensitive instruments which frequently detect non-stationary, non-Gaussian noise transients. iDQ is a statistical inference framework which leverages the use of auxiliary degrees of freedom monitored in the detectors to identify such transients. In this work, we describe the improvements to the iDQ pipeline made between the third and fourth observing run of the LIGO-Virgo-KAGRA (LVK) collaboration, and show the performance of these changes. We find that iDQ detects a total of 39,398 of the known 100,512 glitches identified by Omicron over the course of the second half of the third observing run. We construct a measure of the probability a glitch is present in the strain data of a given detector by combining information from iDQ and Omicron as well as extend the output of iDQ in a novel method which finds correlations between known glitch classifications, and auxiliary channels. We identify several channels over the course of O3b which frequently record instances of Scattered Light, Whistle, and Blip glitches and discuss use cases for this method in active observing runs.

gr-qc↗

arXiv2024

A neural network-based gravitational wave interpolant with applications to low-latency analyses

Matched-filter based gravitational-wave search pipelines identify candidate events within seconds of their arrival on Earth, offering a chance to guide electromagnetic follow-up and observe multi-messenger events. Understanding the detectors' response to an astrophysical transient across the searched signal manifold is paramount to inferring the parameters of the progenitor and deciding which candidates warrant telescope time. We describe a framework that uses artificial neural networks to interpolate gravitational waves and, equivalently, the signal-to noise ratio (SNR) across sufficiently local patches of the signal manifold. Our machine-learning based model generates a single waveform in 6 milliseconds on a CPU and 0.4 milliseconds on a GPU. When using a GPU to generate batches of waveforms simultaneously, we find that we can produce $10^4$ waveforms in $\lesssim 1$ ms. This is achieved while remaining faithful, on average, to 1 part in $10^4$ (1 part in $10^5$) for binary black hole (binary neutron star) waveforms. The model we present is designed to directly utilize intermediate detection pipeline outputs in the hopes of facilitating a better real-time understanding of gravitational-wave candidates.

astro-ph.IM↗

arXiv2024

68-Channel Highly-Integrated Neural Signal Processing PSoC with On-Chip Feature Extraction, Compression, and Hardware Accelerators for Neuroprosthetics in 22nm FDSOI

Multi-channel electrophysiology systems for recording of neuronal activity face significant data throughput limitations, hampering real-time, data-informed experiments. These limitations impact both experimental neurobiology research and next-generation neuroprosthetics. We present a novel solution that leverages the high integration density of 22nm FDSOI CMOS technology to address these challenges. The proposed highly integrated programmable System-on-Chip comprises 68-channel 0.41 \textmu W/Ch recording frontends, spike detectors, 16-channel 0.87-4.39 \textmu W/Ch action potential and 8-channel 0.32 \textmu W/Ch local field potential codecs, as well as a MAC-assisted power-efficient processor operating at 25 MHz (5.19 \textmu W/MHz). The system supports on-chip training processes for compression, training and inference for neural spike sorting. The spike sorting achieves an average accuracy of 91.48% or 94.12% depending on the utilized features. The proposed PSoC is optimized for reduced area (9 mm2) and power. On-chip processing and compression capabilities free up the data bottlenecks in data transmission (up to 91% space saving ratio), and moreover enable a fully autonomous yet flexible processor-driven operation. Combined, these design considerations overcome data-bottlenecks by allowing on-chip feature extraction and subsequent compression.

eess.SP↗

arXiv2023

Improved ranking statistics of the GstLAL inspiral search for compact binary coalescences

Starting from May 2023, the LIGO Scientific, Virgo and KAGRA Collaboration is planning to conduct the fourth observing run with improved detector sensitivities and an expanded detector network including KAGRA. Accordingly, it is vital to optimize the detection algorithm of low-latency search pipelines, increasing their sensitivities to gravitational waves from compact binary coalescences. In this work, we discuss several new features developed for ranking statistics of GstLAL-based inspiral pipeline, which mainly consist of: the signal contamination removal, the bank-$ξ^2$ incorporation, the upgraded $ρ-ξ^2$ signal model and the integration of KAGRA. An injection study demonstrates that these new features improve the pipeline's sensitivity by approximately 15% to 20%, paving the way to further multi-messenger observations during the upcoming observing run.

astro-ph.IM↗

arXiv2017

A geographically distributed bio-hybrid neural network with memristive plasticity

Throughout evolution the brain has mastered the art of processing real-world inputs through networks of interlinked spiking neurons. Synapses have emerged as key elements that, owing to their plasticity, are merging neuron-to-neuron signalling with memory storage and computation. Electronics has made important steps in emulating neurons through neuromorphic circuits and synapses with nanoscale memristors, yet novel applications that interlink them in heterogeneous bio-inspired and bio-hybrid architectures are just beginning to materialise. The use of memristive technologies in brain-inspired architectures for computing or for sensing spiking activity of biological neurons8 are only recent examples, however interlinking brain and electronic neurons through plasticity-driven synaptic elements has remained so far in the realm of the imagination. Here, we demonstrate a bio-hybrid neural network (bNN) where memristors work as "synaptors" between rat neural circuits and VLSI neurons. The two fundamental synaptors, from artificial-to-biological (ABsyn) and from biological-to- artificial (BAsyn), are interconnected over the Internet. The bNN extends across Europe, collapsing spatial boundaries existing in natural brain networks and laying the foundations of a new geographically distributed and evolving architecture: the Internet of Neuro-electronics (IoN).

cs.ET↗

arXiv2017

Analytic saddlepoint approximation for ionization energy loss distributions

We present a saddlepoint approximation for ionization energy loss distributions, valid at arbitrary relativistic velocities of the incident particle 0<v/c<1, provided that ionizing collisions are still the dominant energy loss mechanism. We derive a closed form solution closely related to Moyal's distribution. This distribution is intended for use in simulations with relatively low computational overhead. The approximation generally reproduces the Vavilov most probable energy loss and full width at half maximum to better than 1% and 10%, respectively, with significantly better agreement as Vavilov's kappa approaches 1.

physics.comp-ph↗

arXiv2014

Herschel reveals a molecular outflow in a z = 2.3 ULIRG

We report the results from a 19-h integration with the SPIRE Fourier Transform Spectrometer aboard the Herschel Space Observatory which has revealed the presence of a molecular outflow from the Cosmic Eyelash (SMM J2135-0102) via the detection of blueshifted OH absorption. Detections of several fine-structure emission lines indicate low-excitation HII regions contribute strongly to the [CII] luminosity in this z = 2.3 ULIRG. The OH feature suggests a maximum wind velocity of 700 km/s, which is lower than the expected escape velocity of the host dark matter halo, ~1000 km/s. A large fraction of the available molecular gas could thus be converted into stars via a burst protracted by the resulting gas fountain, until an AGN-driven outflow can eject the remaining gas.

astro-ph.GA↗