Searcharxiv⌕ Search

arXiv subjects

Mihika Sanghi

Publications and source records attributed to Mihika Sanghi.

2 recordsLinked to original sources

Stochastic gravitational wave from graviton bremsstrahlung in inflaton decay into massive spin 3/2 particles

The detection of primordial gravitational waves would offer a direct evidence of inflation and valuable insights into the dynamics of the early universe. During post-inflation reheating period, when the inflaton coherently oscillates at the bottom of its potential, primordial stochastic gravitational waves may be sourced by its perturbative decay into particles of different spins. Assuming the behavior of the potential near the minimum as a polynomial $V(ϕ)\sim ϕ^k$, where $k\ge 2$, and treating the inflaton as coherently oscillating classical field, we calculate the decay of inflaton into a pair of spin $3/2$ particles accompanied by graviton emission. We numerically study the reheating dynamics and calculate the stochastic gravitational wave spectra. Our analysis shows that the gravitational wave spectra can offer insights into the microscopic physics during inflation.

hep-ph↗

Mr. Snuffleupagus at SemEval-2025 Task 4: Unlearning Factual Knowledge from LLMs Using Adaptive RMU

Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language understanding and generation. However, their tendency to memorize training data raises concerns regarding privacy, copyright compliance, and security, particularly in cases involving Personally Identifiable Information (PII). Effective machine unlearning techniques are essential to mitigate these risks, yet existing methods remain underdeveloped for LLMs due to their open-ended output space. In this work, we apply the Adaptive Representation Misdirection Unlearning (RMU) technique to unlearn sensitive information from LLMs. Through extensive experiments, we analyze the effects of unlearning across different decoder layers to determine the most effective regions for sensitive information removal. Our technique ranked 4th on the official leaderboard of both 1B parameter and 7B parameter models.

cs.LG↗