Fullmoonfever
Top 20
Looks like our summer intern last year, posted about by @Frangipani in Nov has an extended stay as well as expanding what he's been up to.
Haven't tried to check the recent publications yet. Just a quick post.
thestockexchange.com.au
Laguna Hills, California, United States
I contributed to the development of TENNs, a novel state-space model optimized for our Spiking Neural Network chip Akida, enabling efficient multimodal processing across audio, text, and vision. I helped develop aTENNuate, a real-time deep state-space speech enhancement model submitted to Interspeech 2025, and explored LoRA-based adaptation for optimizing these models. My work included refining LLM training for efficiency, designing a custom evaluation pipeline, and implementing a Triton-based GPU kernel for FFT convolution to enhance signal processing. Additionally, I developed model obfuscation techniques for secure edge inference and spearheaded a state-space-based speaker verification system for enterprise applications.
Haven't tried to check the recent publications yet. Just a quick post.

BRN Discussion Ongoing
With due respect, no offense intended 1: Really? a former NED venting out his frustration is something you believe unprofessional? He obviously has his skin attached to our beloved BRN, That is why he has every right to express his own opinion with or without inside info 2: We had our director...

FNU Sidharth
ML Research @ BrainChip | Incoming CS PhD @ Univ. of Michigan, Soundability Lab | Speech and Audio Processing | UW ECE
BrainChip University of Washington
Machine Learning Researcher
BrainChip
Jun 2024 - Present 9 monthsLaguna Hills, California, United States
I contributed to the development of TENNs, a novel state-space model optimized for our Spiking Neural Network chip Akida, enabling efficient multimodal processing across audio, text, and vision. I helped develop aTENNuate, a real-time deep state-space speech enhancement model submitted to Interspeech 2025, and explored LoRA-based adaptation for optimizing these models. My work included refining LLM training for efficiency, designing a custom evaluation pipeline, and implementing a Triton-based GPU kernel for FFT convolution to enhance signal processing. Additionally, I developed model obfuscation techniques for secure edge inference and spearheaded a state-space-based speaker verification system for enterprise applications.