gradient-analysis

manage ordinations and render biplots in a tidyverse workflow

Analysis and investigating the confounding effect of accents in end-to-end Automatic Speech Recognition models.

Code for paper: "Improved Residual Network Based on Norm-Preservation for Visual Recognition" https://doi.org/10.1016/j.neunet.2022.10.023

Theorem 3: Selective neurons force Jacobian variation. Verified via Monte Carlo gradient sampling.

A reproducible toolkit that predicts, at design time, the first-order noise susceptibility of gradients in equivariant quantum neural networks, resolved by parameter and noise slot, validated against matched finite-difference derivatives, with engineered zero-response controls, signed coherent cases, and a parameter-resolved validation map.

This repository contains the public release of the Phase-Gradient Cavitation Model. The model is constructed using conventional mathematical tools (fluid analogy, divergence, gradient behavior, and phase-differential structures), and organized using conceptual clarity inspired by the Contextual Premise Framework (CPF).

Empirical comparison of MSE vs Cross-Entropy on MNIST. CE reaches 97% accuracy in 4 epochs vs 13 for MSE - gradient norm analysis confirms the saturation mechanism.

RU: Магистерская диссертация по predictive coding в сопоставимых вычислительных условиях: послойные механизмы, атрибуция стоимости и ограниченное распознавание возможности раннего действия. | EN: MSc thesis on predictive coding under comparable computational conditions: layer-wise mechanisms, cost attribution, and bounded early-action recognition.