singular value decomposition
PulseAugur coverage of singular value decomposition — every cluster mentioning singular value decomposition across labs, papers, and developer communities, ranked by signal.
10 day(s) with sentiment data
-
Deep learning framework optimizes tri-hybrid MIMO precoding for enhanced wireless efficiency
Researchers have developed a novel deep learning framework called Tri-PNet to optimize multi-user MIMO precoding in wireless communication systems. This framework integrates electromagnetic (EM)-reconfigurable antennas …
-
Linguistic theory visualized: Word proximity demo uses PPMI and SVD
This demo illustrates a linguistic theory by visualizing word relationships in a 3D space. It uses Positive Pointwise Mutual Information (PPMI) to quantify how often words appear together more than chance would predict.…
-
New method computes Nash equilibria for cyber-defense games
Researchers have developed a new method called Dynamical Low-Rank Equilibrium Computation (DLR-NE) to efficiently compute Nash equilibria for stochastic games, particularly in the context of industrial control systems (…
-
New research reframes SVD as core ML algorithm, extends to multi-source learning
Two new research papers explore the geometric underpinnings and applications of Singular Value Decomposition (SVD) in machine learning. The first paper re-examines SVD from a geometric perspective, demonstrating how its…
-
New symmetric autoencoders offer theoretical foundation for deep learning
Researchers have introduced a new class of deep learning architectures called symmetric autoencoders, offering a more robust theoretical foundation compared to existing methods. This work formally distinguishes between …
-
New DiDAE method tackles foundation model vulnerabilities with faster counterfactuals
Researchers have introduced Disentangled Diffusion Autoencoders (DiDAE), a novel method designed to address vulnerabilities in foundation models, such as spurious correlations and "Clever Hans" strategies. DiDAE integra…
-
New method learns functional subspaces for efficient neural network compression
Researchers have developed a new method called Learnable Subspace Projections (LSP) to compress large neural networks, particularly transformers and LLMs. Unlike previous techniques that use local criteria, LSP optimize…
-
New framework uses SVD to improve deep learning model fairness
Researchers have developed FairLRF, a novel framework that utilizes singular value decomposition (SVD) to enhance fairness in deep learning models, particularly for sensitive applications like medical diagnosis. Unlike …
-
New framework enhances feature selection with noisy data and relaxed symmetry
Researchers have developed a new framework for universal feature selection that accommodates noisy observations and less restrictive symmetry conditions. This approach, based on the singular value decomposition of a can…
-
WaRA: Wavelet Adaptation for Medical Image Classification
Researchers have introduced WaRA, a novel wavelet-structured adaptation module designed for parameter-efficient fine-tuning of large pretrained vision models in medical image classification. This method operates in a wa…
-
New research offers advanced low-rank compression for LLMs · 3 sources tracked
Three new research papers introduce advanced techniques for compressing large language models (LLMs) using low-rank decomposition. The first paper, 'Per-Matrix Optimality Is Not Enough,' proposes a three-level optimizat…
-
New theory offers lower bound for quantized matrix multiplication error
Researchers have developed a new theoretical framework to minimize error in quantized matrix multiplication. The study, published on arXiv, introduces a nuclear-norm lower bound for dithered scalar quantization, providi…
-
New SVGD framework enhances AI model fine-tuning with geometry awareness
Researchers have developed a new framework for parameter-efficient fine-tuning of large pre-trained models that leverages the geometric structure of low-rank manifolds. This approach utilizes Stein Variational Gradient …
-
New SVDtrunc method significantly compresses Diffusion Transformers for image generation
Researchers have developed a novel method called SVDtrunc for compressing Diffusion Transformers (DiTs), a popular architecture for text-to-image generation. Unlike previous approaches that could lead to performance deg…
-
AI attention mechanisms: SVD compression accelerates rank collapse in pretrained models
A new research paper explores the foundational role of linear algebra in efficient attention mechanisms within AI models. The study unifies fourteen existing works and introduces an original finding: Singular Value Deco…
-
PCST method compresses LLaMA-7B model to 2.05 GiB, but quality lags
Researchers have developed PCST (Product Code Structured Transform), a method for compressing the LLaMA-7B model to 2.05 GiB without retraining. While PCST achieved a smaller file size than the Q3_K_M model, it fell sho…
-
New SVD-MBR method combats overfitting in text generation
Researchers have developed a new method called SVD-MBR to combat overfitting in Minimum Bayes Risk (MBR) decoding for text generation. This technique uses Singular Value Decomposition (SVD) to approximate the utility ma…
-
New frameworks enhance personalized federated learning for LLMs
Two new research papers introduce advanced techniques for personalized federated learning of large language models (LLMs). The first, FedRoRA, addresses rank heterogeneity by decoupling adaptation into shared global dir…
-
New method offers exact error analysis for null-space SVD estimation
Researchers have developed a new method for analyzing errors in null-space estimation from noisy matrices. The study provides exact compact expressions for the error of the smallest left singular vector and all-order se…
-
New methods enhance LoRA efficiency and stability for model adaptation
Researchers have developed two new methods to improve the efficiency and stability of Low-Rank Adaptation (LoRA) techniques used in parameter-efficient model adaptation. Normalized Low-Rank Adaptation (NoRA) normalizes …