Mae
PulseAugur coverage of Mae — every cluster mentioning Mae across labs, papers, and developer communities, ranked by signal.
- authored ScienceCast 90%
- developed SSL4EO 90%
- instance of root-mean-square deviation 70%
- developed by alphaXiv 70%
- developed alphaXiv 70%
- instance of Gotit.pub 70%
- used by alphaXiv 70%
- instance of ScienceCast 70%
- instance of CatalyzeX 70%
- used by Gotit.pub 70%
- used by Dino 70%
- developed Mape Morottaja 70%
11 day(s) with sentiment data
-
LingBot-Touch model bridges tactile data gap for robots
Researchers have developed LingBot-Touch, an open-source tactile pre-training model designed to translate raw sensor data into a format understandable by robot action models. Initial experiments showed that robots equip…
-
AI model fine-tuned for pediatric ECGs shows improved diagnostic accuracy
Researchers have developed a method to adapt adult artificial intelligence models for pediatric electrocardiogram (ECG) interpretation. By fine-tuning an existing adult ECG foundation model with a dataset of 339,563 ped…
-
PointRAFT uses 3D deep learning for accurate potato weight prediction
Researchers have developed PointRAFT, a novel 3D deep learning network designed to accurately predict potato tuber weight from incomplete point cloud data. This method incorporates a dedicated height embedding, leveragi…
-
New method optimizes frozen models for time-series data imputation
Researchers have developed a novel method for imputing missing data in time-series channels by optimizing a frozen sequence reconstructor at inference time. This approach treats missing entries as variables to be optimi…
-
MetaRefine architecture improves image assessment scores using metadata
Researchers have developed MetaRefine, a novel architecture designed to improve image assessment scores by incorporating metadata about the image's original dimensions and composition. This system refines scores by anal…
-
New video fusion network uses dynamic region guidance for efficiency
Researchers have developed an efficient multi-modal video fusion network that improves temporal coherence and detail by dynamically focusing on key regions. This new method avoids computationally expensive optical flow …
-
SpecLST framework enhances traffic forecasting with frequency-domain modeling
Researchers have developed SpecLST, a novel framework designed to improve traffic forecasting by efficiently modeling long-term dependencies and integrating diverse data sources. The system utilizes a fast Fourier trans…
-
New framework enhances wind power forecasting accuracy
Researchers have developed a novel framework called Guarded Agent-in-the-Loop Residual Adaptation to improve wind power forecasting models. This system adapts only a small set of scene-conditioned residual experts, leav…
-
Physics-Aware Self-Supervised Learning Framework for Raman Spectroscopy Unveiled
Researchers have developed a novel self-supervised learning framework for Raman spectroscopy, utilizing a masked autoencoder (MAE) to extract transferable spectral representations from diverse datasets. This approach in…
-
WildMAE enhances wildlife image reconstruction using structure-aware MAE
Researchers have developed WildMAE, a novel extension of the Masked Autoencoder (MAE) framework specifically designed for reconstructing wildlife images. This method enhances MAE by incorporating local structure-aware r…
-
Huber Loss: A Hybrid Approach for Machine Learning
This article delves into the Huber loss function, a hybrid approach that combines the strengths of Mean Squared Error (MSE) and Mean Absolute Error (MAE) for machine learning models. It aims to provide a robust solution…
-
New methods boost Vision Transformer efficiency for image segmentation · 4 sources tracked
Researchers have developed new methods to improve the efficiency of Vision Transformers (ViTs) for image segmentation tasks. TEAL (Token-space Efficient Active Learning) focuses on active learning for semantic segmentat…
-
Progressive depth growth boosts vision transformer performance with less computation
Researchers have explored the benefits of progressive depth growth in self-supervised vision transformers, specifically within the MAE and DINOv3 frameworks. Their findings indicate that this training strategy improves …
-
AI model predicts combined hospital burden for flu, COVID-19, and RSV
Researchers have developed a new framework for forecasting the combined hospital burden of co-circulating respiratory illnesses like influenza, COVID-19, and RSV. This model aims to provide a single planning signal for …
-
Understanding Mean Absolute Error (MAE) in Machine Learning
This article delves into the Mean Absolute Error (MAE) loss function, explaining its properties and applications in machine learning. It discusses how MAE quantifies errors by taking the average of the absolute differen…
-
何恺明团队发布VISTA视觉Harness，提升多模态模型推理能力 · 1 source tracked
Researchers led by何恺明 (Kaiming He) have introduced VISTA, a novel visual harness designed to enhance multimodal models' ability to reason within interactive environments. VISTA allows models to retain and revisit raw vi…
-
New AI pipeline simulates abdominal ultrasounds from semantic labels
Researchers have developed a novel two-stage pipeline for simulating abdominal ultrasound images from semantic labels, bypassing the need for patient-specific CT scans during inference. The first stage uses models like …
-
DSTNet: New AI Model for Causal Multi-Horizon Financial Forecasting
Researchers have developed DSTNet, a novel Dynamic Spectral Trajectory Network designed for causal multi-horizon financial forecasting. Unlike traditional wavelet-based methods, DSTNet retains the recent evolution of fi…
-
New STAMP framework enhances AI's temporal awareness in medical image analysis
Researchers have developed STAMP, a novel Stochastic Siamese Masked Autoencoder framework designed to improve the temporal awareness of AI models in analyzing longitudinal medical images. Unlike deterministic methods, S…
-
New time series foundation models tackle zero-shot forecasting and scale invariance
Researchers have introduced Timer-M1, a novel multivariate time series foundation model designed for zero-shot forecasting by learning temporal and relational primitives. This model utilizes a primitive-based data synth…