ViT-B/16
PulseAugur coverage of ViT-B/16 — every cluster mentioning ViT-B/16 across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
-
Pretraining and Distillation Outperform Architecture Choice in Cell Classification
A new study published on arXiv investigates the effectiveness of different deep learning architectures for label-free single-cell classification. The research found that pretraining and fine-tuning strategies are more c…
-
New MCANet model improves post-hurricane damage assessment from UAV imagery
Researchers have developed MCANet, a novel multi-label classification framework designed for assessing post-hurricane damage using UAV imagery. This network integrates a Res2Net backbone for multi-scale feature extracti…
-
CLIP models can suffer performance loss with larger text encoders
A new arXiv paper reveals that increasing the size of text encoders in CLIP models can negatively impact zero-shot performance, even when the total parameter count increases. Researchers found that for most vision encod…
-
UC Berkeley researchers develop bandit-based pruning for transformers
Researchers from the University of California, Berkeley have developed a novel method for pruning large transformer models, including those used in vision and language tasks. This technique, framed as a damage-aware mul…
-
New method tackles rotation-induced drift in AI model explanations
Researchers have identified a significant issue with post-hoc saliency maps, such as Grad-CAM, used to audit AI model decisions. These maps exhibit a 'drift' when input images are rotated, even if the model's prediction…
-
New research explores advanced machine unlearning techniques for AI models · 4 sources tracked
Researchers are developing new methods for machine unlearning, the process of removing specific data or knowledge from AI models. One approach, Source-Free Class Relearning Audit (SFRA), focuses on recovering forgotten …
-
New method enables efficient fine-tuning of ternary transformers
Researchers have developed a new method called ternary multiplicative adaptation for fine-tuning transformers that are quantized to ternary weights. This approach uses a low-rank Kronecker factorization to represent dis…
-
New audit method reveals true interactions in language models
Researchers have developed a new method called a site-asymmetry audit to better understand interactions within neural networks. This audit helps distinguish genuine interaction effects from those caused by the location …
-
New method uses SVD to detect out-of-distribution data in Vision Transformers
Researchers have developed a novel method for detecting out-of-distribution (OOD) data in Vision Transformers (ViTs) by analyzing the geometry of their learned parameters. The approach involves factoring each affine lay…
-
New benchmark reveals when AI knowledge fusion helps or harms
A new benchmark study explores the effectiveness of combining hand-crafted knowledge with learned representations in AI models, particularly when training data is scarce. The research found that different types of knowl…
-
SPARCL method tackles spectral interference in analytic continual learning
Researchers have introduced SPARCL, a novel analytic continual learning method that addresses the issue of spectral interference in existing approaches. Unlike previous methods that suffer from forgetting old classes du…
-
AI pipeline unlocks recognition of ancient Elamite cuneiform symbols
Researchers have developed EpigraphNet, a novel pipeline for recognizing Elamite cuneiform symbols from degraded tablet images. This system utilizes zero-shot SAM2 segmentation to create clean symbol masks, which are th…
-
New methods improve industrial anomaly detection calibration and evaluation
Researchers have developed new methods for calibrating industrial anomaly detection systems, particularly when faced with distribution shifts and a scarcity of labeled anomalies. One approach, SPARC, uses a few verified…
-
DCA-MoE framework enhances crowd counting with adaptive fusion and routing
Researchers have introduced DCA-MoE, a novel framework designed to improve crowd counting accuracy by making feature fusion and expert routing content-dependent. This approach utilizes Spatially Adaptive Layer Fusion (S…
-
New dataset and models advance sign language handshape recognition
Researchers have developed a new dataset and baseline models for fine-grained isolated handshape recognition in sign language, utilizing the HamNoSys notation system. The dataset comprises 144,000 RGB images from 15 par…
-
New framework audits AI face analysis for hidden fairness risks
Researchers have developed a new framework called CIFA (Contextual-Intersectional Fairness Auditing) to identify hidden vulnerabilities in face analysis systems. This framework goes beyond traditional demographic fairne…
-
New framework uncovers hidden fairness flaws in face analysis AI
Researchers have developed a new framework called CIFA (Contextual-Intersectional Fairness Auditing) to identify hidden vulnerabilities in computer vision models, particularly in face analysis. This framework goes beyon…
-
New AI Attribution Method Boosts Robustness with Minimal Accuracy Loss
Researchers have developed a new framework to improve the faithfulness and consistency of attribution methods in AI models, particularly under geometric transformations. This annotation-free approach uses submodular sea…
-
New SpecTraL method improves federated LoRA for Vision Transformers
Researchers have developed a new method called SpecTraL for improving federated learning of Vision Transformers (ViTs) using low-rank adapters (LoRA). This approach addresses limitations in existing strategies, such as …
-
AI benchmark leakage identified, impacting OOD detection accuracy
Researchers have identified a significant issue with benchmark datasets used for evaluating out-of-distribution (OOD) detection in AI models. They discovered that some benchmarks contain data from the model's training s…