ONM Student Talk: Ram Gaurav @ Virginia Tech

Ramashish Gaurav (Ram) is a 3rd year Ph.D. student at Virginia Tech, USA. He is supervised by Prof. Yang (Cindy) Yi in her BRICC Lab, ECE @ VT. Of late, Ram has been working on reservoir-based spiking models for Time Series Classification (TSC). Reservoir Computing is a well-established domain for time-series processing where a reservoir of statically (and recurrently) connected neurons compute high-dimensional temporal features, over which a linear readout layer learns the mapping to the output.

In his recent work [1], Ram designed the Legendre-SNN (LSNN), a simple – yet high performing SNN model (for univariate TSC) where he has used the Legendre Delay Network (LDN) [2] as a non-spiking reservoir (in fact, the LDN in LSNN is implemented with just basic matrix-operations). In a subsequent work (currently under review), he extended his LSNN to DeepLSNN that accounts for multivariate time-series signals too; upon experimenting with it, he found that DeepLSNN models outperform a popular (and complex) LSTM-Conv integrated model [3] on more than 30% of 101 TSC datasets. His latest work is on the evaluation of Legendre-SNN on the Loihi-2 chip [4] — on which this talk is focused at.

Talk details here, time TBA.