Showing cs.LGShow all
2 papers · 1 filter
cs.LG2026
Argus: A Real-EKS Study of When Predicting Spot Interruptions Beats Simple Checkpointing
Angshuman Chakravertty, MD Rayyan
Elastic Compute Cloud (EC2) Spot is 60% to 90% cheaper than On-Demand but can be reclaimed on just a 2-minute notice; for expensive multi-node training this loss can be severe, wit…
cs.LG2026
VESTIGE: A Knowledge-Guided Masking Strategy for Corruption-Aware Fine-Tuning of Genomic Transformers, Validated on Ancient DNA Reconstruction
Angshuman Chakravertty, Rahul Maheshwari
Standard masked-language-model fine-tuning applies a uniform masking probability across every token position, assuming reconstruction difficulty is position-agnostic. When the degr…