Imagenet 1k
PulseAugur coverage of Imagenet 1k — every cluster mentioning Imagenet 1k across labs, papers, and developer communities, ranked by signal.
6 day(s) with sentiment data
-
New Fusion Method Merges Dissimilar Vision Models
Researchers have developed a novel method called Riemannian--Lorentz Parameter Fusion (RLPF) to merge independently trained vision models, even when their architectures differ. This technique addresses the challenges of…
-
New CTOAC method improves low-bit quantization for Visual State Space Duality models
Researchers have developed a new post-training quantization (PTQ) method called Channel-wise Token-balanced Output-Aware Clipping (CTOAC) to address the low-bit quantization challenges in Visual State Space Duality (VSS…
-
New method generates domain-specific datasets on-demand, outperforming large general datasets
Researchers have developed a novel method called Precision at Scale (PaS) for automatically generating domain-specific datasets on demand. This approach challenges the conventional wisdom that massive, general-domain da…
-
Self-supervised models show strong generalization to unseen datasets
Researchers have conducted an empirical study to evaluate how well self-supervised encoders generalize to unseen datasets without retraining. The study deployed image models pretrained on ImageNet-1k, comparing supervis…
-
New OOD benchmark Fi-ImageNet-1k challenges AI models
Researchers have developed Fi-ImageNet-1k, a new dataset designed to challenge out-of-distribution (OOD) detection capabilities in AI models. This dataset is derived from ImageNet-1k validation images that were reannota…
-
New CORD adapter preserves top-1 predictions in post-hoc calibration
Researchers have introduced CORD (Calibrator-Output Repair for Top-1 Decision Preservation), a novel post-fit adapter designed to improve post-hoc calibration in machine learning models. CORD ensures that while confiden…
-
Image augmentation techniques tested as generators for deep learning image retrieval systems
This paper introduces a novel approach to testing deep learning-based image retrieval systems by utilizing image augmentation techniques as test generators. The research categorizes 50 augmentation methods and empirical…
-
Iwin Transformer tackles ViT complexity with interleaved windows and convolution
Researchers have introduced the Iwin Transformer, a novel hierarchical vision transformer designed to overcome limitations in existing Vision Transformers (ViTs). This new architecture combines interleaved window attent…
-
New Mixture of Channel Experts layer slashes CNN computation
Researchers have introduced Mixture of Channel Experts (MoCE), a novel layer designed to replace standard pointwise projections in convolutional neural networks. Unlike traditional Mixture-of-Experts (MoE) models that s…
-
New video pretraining method and 10M-hour dataset released
Researchers have introduced LeVJEPA, a novel video pretraining method that significantly reduces computational costs while maintaining or improving downstream accuracy. This approach bypasses common heuristics like arch…
-
New Mixture of Channel Experts method boosts CNN efficiency
Researchers have developed a new method called Mixture of Channel Experts (MoCE) to improve the efficiency of convolutional neural networks. Unlike traditional Mixture-of-Experts models that route inputs through differe…
-
New Multi-Overlapped-Head Self-Attention boosts Vision Transformer performance
Researchers have introduced Multi-Overlapped-Head Self-Attention (MOHSA), a novel mechanism designed to enhance Vision Transformers. Unlike standard Multi-Head Self-Attention (MHSA) which isolates attention heads, MOHSA…
-
New research tackles LLM quantization robustness and uncertainty preservation
Two new research papers explore methods to improve the quantization of large language models (LLMs) while preserving their performance and uncertainty behavior. The first paper, "Jacobian-guided Noise Injection for Quan…
-
New method improves deep learning with stable matrix log normalization
Researchers have developed a novel method for matrix logarithm normalization in Global Covariance Pooling (GCP) for deep learning models. This new approach uses orthogonal polynomial approximations, specifically a degre…
-
Wavelet bases beyond Haar offer efficiency gains in AI models
Researchers have explored the effectiveness of different wavelet bases beyond the commonly used Haar and Daubechies for wavelet convolution layers. Their study, focusing on the trade-off between filter length and decomp…
-
ImageNet-1k dataset re-annotated, revealing 12% label errors
A new paper details the extensive process of re-annotating the ImageNet-1k dataset, revealing significant issues with the original labels. The researchers found that approximately 12% of labels were incorrect, 33.3% of …
-
SSOG-Attention offers scalable alternative to SDPA with reduced complexity
Researchers have developed SSOG-Attention, a novel approach that offers a sub-quadratic and scalable alternative to standard Scaled Dot-Product Attention (SDPA). By learning Gaussian atoms and geometrically steering the…
-
New DECAF method offers deeper insights into AI model perturbation responses
Researchers have developed a new method called DECAF (Decomposition of Evidence, Contradiction, And Fragility) to better understand how AI models respond to perturbations. Unlike traditional methods that only measure th…
-
New methods enhance Spiking Transformer performance on image tasks · 2 sources tracked
Researchers have introduced Spiking Local Interaction (SLI) and Adaptive Complementary Fusion (ACF) to enhance Spiking Transformers. These methods address limitations in standard Spiking Self-Attention (SSA) by introduc…
-
New methods accelerate Vision Transformer adaptation for edge devices
Researchers have developed new methods for adapting Vision Transformers (ViTs) to specific tasks more efficiently. One approach uses genetic programming to evolve layer-specific scalar functions that approximate normali…