BERKELEY M.E.T. / EECS

Sachin
Mahesh.

Exploring how intelligent systems work.
Making a little music along the way.

RESEARCH · ENGINEERING · MUSICBERKELEY, CA

01 / RESEARCHMANUSCRIPT

Control,
not content.

Input-Dependent Weight Generation
as a Neural Control Plane

What changes when a neural network generates its own weights? My research examines how these updates steer behavior—and where they fall short as a way to carry information.

Read the paper

THE CENTRAL DISTINCTION

Control plane

Generated weights shape
how a network behaves.

ΔW(c) = B diag(g(c)) A

Content memory

Recurrent state retains
what a network knows.

Complementary roles. Different mechanisms.

HypernetworksAdaptive architecturesMultimodal learning

02 / MUSICANOTHER SIDE OF ME

Off the page.
Into sound.

I also make music as silverthrone.

Listen on Spotify

silverthrone

Original music · Open in Spotify ↗

03 / ABOUT

Curiosity across
disciplines.

I'm Sachin, studying EECS in UC Berkeley’s Management, Entrepreneurship, & Technology program. My interests span intelligent systems, engineering, and music.

This is a home for the work I’m thinking through and the things I’m making.

Let’s connect on LinkedIn

SachinMahesh.

Control,not content.