The Street View House Numbers Dataset
PulseAugur coverage of The Street View House Numbers Dataset — every cluster mentioning The Street View House Numbers Dataset across labs, papers, and developer communities, ranked by signal.
8 day(s) with sentiment data
-
Canonical JEM models show indistinguishable performance between PC and SGLD samplers
Researchers have investigated the performance of two sampling methods, Predictor-Corrector (PC) and Stochastic Gradient Langevin Dynamics (SGLD), when applied to Canonical Joint Energy-Based Models (JEM) on the CIFAR-10…
-
New AI unlearning method irreversibly erases data
Researchers have developed a new machine unlearning method called One-Point Contraction (OPC) that aims to irreversibly erase data from AI models. Unlike existing methods that merely obscure information, OPC collapses f…
-
AI advances photonic component design with neurosymbolic and BNN approaches
Researchers have developed new methods for designing photonic components using AI. One approach, "Constrained Co-Design for Photonic Bayesian Neural Networks," focuses on improving the uncertainty estimation of Bayesian…
-
EulerLoRA enhances parameter-efficient fine-tuning with stochasticity
Researchers have developed EulerLoRA, a novel extension of the Low-Rank Adaptation (LoRA) technique for parameter-efficient fine-tuning. Unlike standard LoRA, EulerLoRA introduces stochasticity to generate multiple pred…
-
New framework enhances Deep JSCC for image classification across domains
Researchers have developed a novel domain-adaptive framework for Deep Joint Source-Channel Coding (Deep JSCC) to improve image classification performance under distribution shifts. The proposed Classification-Capacity-I…
-
AI research uses multi-armed bandits to prune neural networks
Researchers have developed a novel method for pruning feature maps in convolutional neural networks (CNNs) to reduce computational costs and storage requirements. This approach utilizes multi-armed bandit algorithms, sp…
-
New methods enhance differential privacy in deep neural network training · 2 sources tracked
Two new research papers propose novel methods for training deep neural networks with differential privacy, aiming to improve both accuracy and efficiency. The first paper introduces an end-to-end framework that privatiz…
-
KALE method improves CLIP visual representations using adaptive loss equilibration
Researchers have developed KALE (Kernel Alignment with Loss Equilibration), a novel method to improve CLIP's visual representations by aligning it with a vision-centric teacher model like DINOv2. Unlike previous approac…
-
New research reveals temperature scaling distorts AI model error proxies
A new paper published on arXiv details how temperature scaling, a common post-hoc calibration method for AI models, can significantly distort Bayes-error proxy estimates. Researchers Ishida and Ushio found that this dis…
-
LLMs fine-tuned with MinHash curriculum for neural architecture synthesis
Researchers have developed a novel framework for neural architecture search (NAS) that utilizes a MinHash-based similarity scheduling approach to create a progressive curriculum for fine-tuning large language models (LL…
-
New LDPKiT Framework Enhances Privacy in Model Distillation
Researchers have developed LDPKiT, a novel framework designed for privacy-preserving model distillation. This method allows users to leverage a model's capabilities using their own private data while bounding privacy le…
-
Quantum GAN generates full-resolution images without tricks
Researchers have developed a novel quantum generative adversarial network (qGAN) capable of generating full-resolution images from classical datasets like MNIST and Fashion-MNIST. This approach circumvents the need for …
-
New frameworks enhance Federated Learning privacy, robustness, and efficiency · 4 sources tracked
Researchers are developing advanced frameworks for Federated Learning (FL) to enhance privacy, robustness, and efficiency. PRoVeFL utilizes multi-key fully homomorphic encryption across multiple servers to protect again…
-
LLMs guide neural network generation, improving accuracy via source-model guidance · 2 sources tracked
Researchers have developed a novel protocol for using large language models (LLMs) to improve existing neural networks by guiding the generation process with a stronger, same-family source model. This method aims to dis…
-
New theory explains how pretraining shapes machine learning model fine-tuning
Researchers have developed a theoretical framework to explain how pretraining influences inductive bias during the fine-tuning of machine learning models. Their analysis, conducted on diagonal linear networks, identifie…
-
New 'Pre-Warm' method improves CNN initialization accuracy
Researchers have developed a novel method called Pre-Warm for initializing convolutional neural networks. This technique conditions the initialization of the first convolutional layer using data from a single training b…
-
New research questions validity of AI feature attribution benchmark
A new paper from Junghoon Seo on arXiv explores the limitations of the RemOve-And-Retrain (ROAR) benchmark, commonly used to assess feature attribution methods. The research indicates that post-processing attribution ma…
-
Optimized optics boost AI classification under detector limits
Researchers have developed a theoretical framework to understand when optimizing optical front-ends with neural network back-ends improves imaging classification performance. The study found that these gains are most si…
-
New method predicts neural network generalization using Fourier fractal dimension
Researchers have developed a new method to predict how well deep neural networks will generalize without needing separate validation data. This approach uses the Fourier fractal dimension of the network's weight variati…
-
New research probes test-time adaptation challenges in accuracy and latency
Three new research papers explore the nuances of test-time adaptation (TTA) in machine learning. One paper investigates the trade-off between recognizing in-distribution data and detecting out-of-distribution data, find…