spectral-analysis

Tag

Cards List
#spectral-analysis

Spectral Outliers Reveal Dominant Learned Structure in Transformer Attention

arXiv cs.LG · 2026-08-11 Cached

This paper applies Marchenko-Pastur random matrix theory to pre-trained attention weights, separating each projection matrix into a random-like bulk and spectral outliers. Causal experiments show zeroing these outliers in Mistral-7B drives performance near random chance, revealing that spectral outliers encode dominant learned structure across 11 transformers.

0 favorites 0 likes
#spectral-analysis

Phase Structure in Rotary Attention: A Spectral Framework for Semantic Continuity and Execution-Boundary Governance

arXiv cs.CL · 2026-07-29 Cached

This paper develops a spectral framework for analyzing rotary phase alignment, semantic continuity, and representation drift in transformer language models, proposing a bounded spectral method and distinguishing internal coherence from execution-boundary governance.

0 favorites 0 likes
#spectral-analysis

Preference Tuning as Spectral Update Reorganization

arXiv cs.CL · 2026-07-24 Cached

The paper reveals that preference-based post-training induces parameter updates with a spectral head-tail organization, where a compact head carries the dominant behavioral shift and a weak tail is necessary for full solution recovery, recasting alignment as structured update reorganization rather than monolithic correction.

0 favorites 0 likes
#spectral-analysis

Detecting Neural Network Failures through Spectral Analysis of Internal Activations

arXiv cs.LG · 2026-07-24 Cached

This paper identifies spectral drift in internal activations of neural networks during misclassifications and introduces Self-Detecting Neural Networks (SDNN) that monitors spectral dynamics to detect failures, achieving 79% AUROC on CIFAR-10, outperforming confidence-based methods by 25-30 percentage points.

0 favorites 0 likes
#spectral-analysis

The Devil is in the Spectrum: Mitigating Representation Collapse in LLMs via Topologically Regularized Side-Path

arXiv cs.AI · 2026-07-24 Cached

Proposes Topologically Regularized Side-Path (TRSP) to mitigate representation collapse in LLMs by balancing spectral trade-offs between mixing efficiency and information capacity, achieving significant gains on long-context benchmarks.

0 favorites 0 likes
#spectral-analysis

Recurrent Sinusoidal INRs for Efficient High-Fidelity Representation

Hugging Face Daily Papers · 2026-07-23 Cached

This paper introduces a recurrent sinusoidal architecture for implicit neural representations that achieves higher fidelity with fewer parameters and optimization steps by exploiting harmonic line spectrum enrichment through sinusoidal recurrence.

0 favorites 0 likes
#spectral-analysis

Beyond Output-Space Calibration: Spectral Evidence Bundling for Selective Reliability Estimation in Time-Series Classification

arXiv cs.LG · 2026-07-22 Cached

This paper introduces SEB-Cal, a method that augments output-space calibration with spectral features (band energy, entropy, peak dominance, phase stability) to improve selective reliability estimation in time-series classification, achieving higher Corr-AUROC and lower [email protected] across multiple datasets.

0 favorites 0 likes
#spectral-analysis

How the Hessian-Spectrum of Neural Networks Depends on Data

arXiv cs.LG · 2026-07-16 Cached

This paper derives the eigenvalues of the Hessian for linear neural networks of arbitrary width and depth, showing that sharpness relates to maximum class proportion for classification tasks with MSE loss, and empirically validates the predictions.

0 favorites 0 likes
#spectral-analysis

Spectral Origins of the Self-Correction Blind Spot in Autoregressive Generation

arXiv cs.LG · 2026-07-14 Cached

This paper introduces SPARC, a spectral-algebraic theory explaining the self-correction blind spot in autoregressive language models, where models fail to correct their own errors but can fix identical external errors. The theory proves the blind spot arises when the spectral radius of an error-propagation operator is at least one, derives a threshold for correction markers, and provides convergence guarantees for RL-based self-correction training.

0 favorites 0 likes
#spectral-analysis

Complexity-Guided Component-wise Initialization for Language Model Pretraining

arXiv cs.CL · 2026-07-13 Cached

This paper analyzes spectral patterns in pretrained GPT-2-style language models and tests whether these patterns can be used for initialization, finding that coarse spectral matching does not improve pretraining performance over standard methods.

0 favorites 0 likes
#spectral-analysis

Fingerprint, Not Blueprint: How Positional Schemes Set the Default Spectral Algebra of Attention

arXiv cs.LG · 2026-07-09 Cached

This paper investigates the spectral properties of the QK operator in attention heads, showing that the positional scheme (RoPE, learned absolute, ALiBi) sets a default spectral algebra that acts as a fingerprint consolidated after function rather than a hard constraint.

0 favorites 0 likes
#spectral-analysis

Spectral Rewiring for Exploration, Purification, and Model Merging

arXiv cs.LG · 2026-07-07 Cached

The paper introduces Subspace-Aligned Rewiring (SAR), a post-hoc editing method that retains the spectral core of RL updates to preserve reasoning gains, remove interference, and enable model merging across experts, achieving strong performance with minimal parameters.

0 favorites 0 likes
#spectral-analysis

Geometric Signatures of Reasoning: A Spectral Perspective on Task Hardness

arXiv cs.LG · 2026-07-03 Cached

This paper studies the geometric properties of chain-of-thought trajectories in the hidden state space of transformers, introducing effective dimension and kinematic features to predict task hardness and solution correctness from early tokens.

0 favorites 0 likes
#spectral-analysis

Black-Box Inference of LLM Architectural Properties with Restrictive API Access

arXiv cs.LG · 2026-07-03 Cached

This paper presents NightVision, an attack that uses restrictive black-box API access to estimate hidden dimension, depth, and parameter count of large language models. It exploits a novel common-set prompting technique and spectral analysis, achieving high accuracy on open-source models.

0 favorites 0 likes
#spectral-analysis

Multi-Adapter PPO: A Cross-Attention Enhanced Wavelength Selection Framework for LIBS Quantitative Analysis

arXiv cs.LG · 2026-06-17 Cached

This paper introduces Multi-Adapter PPO, a reinforcement learning framework with cross-attention for wavelength selection in LIBS quantitative analysis, achieving 28.4% better composite scores and 45.2% improvement in prediction accuracy over traditional methods on steel and coal datasets.

0 favorites 0 likes
#spectral-analysis

@teropa: I which @sedielem beautifully illustrates why diffusion models work so well with images Our visual world is spatially c…

X AI KOLs Following · 2026-06-16 Cached

An explanation of why diffusion models work well for images: low-frequency spectral components dominate, so denoising recovers coarse structure first, then fine detail — analogous to spectral autoregression.

0 favorites 0 likes
#spectral-analysis

Multi-Scale Feature Attention Network for Polymer Classification using THz Dual-Comb Spectroscopy

arXiv cs.LG · 2026-06-08 Cached

This paper introduces the Multi-Scale Feature Attention Network (MSFAN), a deep learning architecture for classifying 12 types of polymers using THz Dual-Comb Spectroscopy, achieving 85.2% accuracy and outperforming state-of-the-art models.

0 favorites 0 likes
#spectral-analysis

Show HN: Resonate – Low-latency, high-resolution spectral analysis

Hacker News Top · 2026-06-06 Cached

Resonate is a low-latency, low-memory algorithm for perceptually relevant spectral analysis of audio signals, using resonator models with exponentially weighted moving averages.

0 favorites 0 likes
#spectral-analysis

Spectral Asymptotics of Neural Network Loss Landscapes: An Exact Decomposition of the Curvature Exponent

arXiv cs.LG · 2026-06-03 Cached

This paper presents an exact decomposition of the curvature exponent α in neural network loss landscapes, explaining why it varies across layer types. It introduces the spectral alignment decomposition and derives a spectral transfer identity linking curvature, gradient rank decay, and Hessian exponents, validated across architectures and datasets.

0 favorites 0 likes
#spectral-analysis

BitsMoE: Efficient Spectral Energy-Guided Bit Allocation for MoE LLM Quantization

arXiv cs.LG · 2026-06-02 Cached

BitsMoE introduces a spectral-energy-guided bit allocation framework for quantizing Mixture-of-Experts LLMs, achieving substantial accuracy improvements and speedups under ultra-low-bit quantization.

0 favorites 0 likes
Next →
← Back to home

Submit Feedback