3 papers
cs.CL2026
ARCHead: Activation-Metric Residual Correction for Large Language Model Output Heads
Åuayp Talha Kocabay, Şuayp Talha Kocabay, Talha Rüzgar AkkuÅ +2
Weight-only quantization substantially reduces the storage of large language model (LLM) transformer blocks, but practical backends often retain the final language-modeling head (L…
cs.LG2026
Sample Complexity of Scientific Discovery: PAC Learnability of Compositional Function Trees
Åuayp Talha Kocabay, Talha Rüzgar AkkuÅ, Kerem Yalçın
Scientific discovery via symbolic regression is often viewed as statistically and computationally intractable because the hypothesis space of expressions grows combinatorially with…
cs.LG2026
Superpositional Gradient Descent: Harnessing Quantum Principles for Model Training
Ahmet Erdem Pamuk, Emir Kaan Ãzdemir, Åuayp Talha Kocabay
Large language models (LLMs) are increasingly trained with classical optimization techniques like AdamW to improve convergence and generalization. However, the mechanisms by which…