Pretraining and alignment induce asymmetric geometric traces in transformer weights because alignment updates concentrate in read pathways due to activation covariance while write pathways inherit less structure from alignment losses.
Ash, and Dipendra Misra
9 Pith papers cite this work. Polarity classification is still indexing.
9
Pith papers citing it
citation-role summary
method 1
citation-polarity summary
roles
method 1polarities
use method 1representative citing papers
Transformer activations show spectral anti-concentration for concepts in the tail while syntax prefers high-variance directions, forming a dual geometry.
Uncertainty and correctness in LLMs are encoded by distinct feature populations, with suppression of confounded features improving accuracy and reducing entropy.
Holmes is a probing benchmark compiling over 200 datasets from 270 studies to evaluate linguistic competence across syntax, morphology, semantics, reasoning, and discourse in more than 50 language models.
DG-Hard uses Donoho-Gavish hard thresholding on the fine-tuning weight delta to separate task-aligned signal from noise-like residual, recovering damaged capabilities while preserving target-task gains.
BSI ranks singular-vector bases for LLM low-rank compression by estimating expected task loss increase via second-order Taylor expansion of the loss and an efficient Hessian-diagonal estimator, outperforming magnitude-based baselines on math reasoning benchmarks.
DASH-KV accelerates long-context LLM inference to linear complexity via asymmetric KV cache hashing and mixed-precision retention, matching full attention performance on LongBench.
Gradient-guided layer selection for LoRA yields 15-28% training speedup with matched downstream results on MMLU, GSM8K, and HumanEval across 14 models from 0.5B to 72B parameters.
HTMuon modifies Muon to produce heavier-tailed updates and weight spectra via HT-SR theory, yielding up to 0.98 lower perplexity on LLaMA pretraining and serving as a plug-in for other Muon variants.
citing papers explorer
-
Where Pretraining writes and Alignment reads: the asymmetry of Transformer weight space
Pretraining and alignment induce asymmetric geometric traces in transformer weights because alignment updates concentrate in read pathways due to activation covariance while write pathways inherit less structure from alignment losses.
-
Concepts Whisper While Syntax Shouts: Spectral Anti-Concentration and the Dual Geometry of Transformer Representations
Transformer activations show spectral anti-concentration for concepts in the tail while syntax prefers high-variance directions, forming a dual geometry.
-
Are LLM Uncertainty and Correctness Encoded by the Same Features? A Functional Dissociation via Sparse Autoencoders
Uncertainty and correctness in LLMs are encoded by distinct feature populations, with suppression of confounded features improving accuracy and reducing entropy.
-
Holmes: A Benchmark to Assess the Linguistic Competence of Language Models
Holmes is a probing benchmark compiling over 200 datasets from 270 studies to evaluate linguistic competence across syntax, morphology, semantics, reasoning, and discourse in more than 50 language models.
-
Spectral Unforgetting: Post-Hoc Recovery of Damaged Capabilities Without Retraining
DG-Hard uses Donoho-Gavish hard thresholding on the fine-tuning weight delta to separate task-aligned signal from noise-like residual, recovering damaged capabilities while preserving target-task gains.
-
Importance-Guided Basis Selection for Low-Rank Decomposition of Large Language Models
BSI ranks singular-vector bases for LLM low-rank compression by estimating expected task loss increase via second-order Taylor expansion of the loss and an efficient Hessian-diagonal estimator, outperforming magnitude-based baselines on math reasoning benchmarks.
-
DASH-KV: Accelerating Long-Context LLM Inference via Asymmetric KV Cache Hashing
DASH-KV accelerates long-context LLM inference to linear complexity via asymmetric KV cache hashing and mixed-precision retention, matching full attention performance on LongBench.
-
Aletheia: Gradient-Guided Layer Selection for Efficient LoRA Fine-Tuning Across Architectures
Gradient-guided layer selection for LoRA yields 15-28% training speedup with matched downstream results on MMLU, GSM8K, and HumanEval across 14 models from 0.5B to 72B parameters.
-
HTMuon: Improving Muon via Heavy-Tailed Spectral Correction
HTMuon modifies Muon to produce heavier-tailed updates and weight spectra via HT-SR theory, yielding up to 0.98 lower perplexity on LLaMA pretraining and serving as a plug-in for other Muon variants.