# Masked Autoencoder

AudioSphere extends self-supervised audio pretraining from single-channel to multichannel: it takes four-channel Ambisonics recordings and trains a masked autoencoder to reconstruct both spectrograms and intensity vectors, so a frozen backbone with a shallow probe performs sound event detection and localization in real scenes. Prior single-channel models were blind to spatial structure, while spatial models relied on clip-level supervision. It reaches an aggregate HEAR score of 83.2 and also supports frame-level localization.

AudioSphere extends self-supervised audio pretraining from single-channel to multichannel: it takes four-channel Ambisonics recordings and trains a masked autoencoder to reconstruct both spectrograms and intensity vectors, so a frozen backbone with a shallow probe performs sound event detection and localization in real scenes. Prior single-channel models were blind to spatial structure, while spatial models relied on clip-level supervision. It reaches an aggregate HEAR score of 83.2 and also supports frame-level localization.

Radar 3D perception has long compressed raw spectra into sparse point clouds or compact Doppler statistics before supervised training. L-SPMAE instead uses masked self-supervision: whatever remains reconstructable under localized corruption counts as signal, and spectral leakage supplies that corruption, so pretraining runs directly on unreduced radar spectra. On K-Radar, L-SPMAE with the RAD representation reaches 55.9/50.1 mAP after fine-tuning, above the best point-cloud self-supervised baseline in the table at 50.8/45.0.

Radar 3D perception has long compressed raw spectra into sparse point clouds or compact Doppler statistics before supervised training. L-SPMAE instead uses masked self-supervision: whatever remains reconstructable under localized corruption counts as signal, and spectral leakage supplies that corruption, so pretraining runs directly on unreduced radar spectra. On K-Radar, L-SPMAE with the RAD representation reaches 55.9/50.1 mAP after fine-tuning, above the best point-cloud self-supervised baseline in the table at 50.8/45.0.

PlacePulse compresses 1,055,537 US H3 resolution-7 cells into 256-dimensional unit-norm vectors using only static structural attributes such as demographics, businesses and environmental risk, beating a dimensionality-matched feature-selection baseline on 74 external targets and leading Google's Population Dynamics Foundation Model on health and economic outcomes.

PlacePulse compresses 1,055,537 US H3 resolution-7 cells into 256-dimensional unit-norm vectors using only static structural attributes such as demographics, businesses and environmental risk, beating a dimensionality-matched feature-selection baseline on 74 external targets and leading Google's Population Dynamics Foundation Model on health and economic outcomes.

The study introduces TotalFM, an organ-separated 3D-CT radiology foundation model that uses TotalSegmentator and LLMs to automatically build roughly 340,000 organ-level volume-text pairs, then combines VideoMAE self-supervised pre-training with organ-wise contrastive learning; it reaches an average F1 of 0.708 in zero-shot organ-wise lesion classification (versus 0.515 for CT-CLIP and 0.650 for Merlin), a higher AUROC than Merlin in 83% (25/30) of finding categories, and report-generation performance comparable to Merlin while raising batch efficiency to 32 batch/GPU.

The study introduces TotalFM, an organ-separated 3D-CT radiology foundation model that uses TotalSegmentator and LLMs to automatically build roughly 340,000 organ-level volume-text pairs, then combines VideoMAE self-supervised pre-training with organ-wise contrastive learning; it reaches an average F1 of 0.708 in zero-shot organ-wise lesion classification (versus 0.515 for CT-CLIP and 0.650 for Merlin), a higher AUROC than Merlin in 83% (25/30) of finding categories, and report-generation performance comparable to Merlin while raising batch efficiency to 32 batch/GPU.

The study proposes PIRTA, a retrieval-augmented generation framework in which a 3D ViT encoder pretrained with large-scale MAE self-supervision retrieves clinically similar 3D DWI/ADC volumes in the image domain, and their paired clinician-authored reports ground LLaMA3-8B-Instruct report generation, avoiding explicit image-text alignment; on a 1,831-case multi-institutional internal cohort, a 580-case privacy-preserving external cohort, and the 206-case public ISLES benchmark, PIRTA achieves strong image-domain retrieval (internal mAP@1 of 94.0%) and consistently improves ischemic-territory accuracy, a clinically grounded surrogate for factuality, over direct image-to-text 2D baselines, leading the strongest 2D baseline by +57.2 / +34.5 / +30.6 multi-class accuracy points.

The study proposes PIRTA, a retrieval-augmented generation framework in which a 3D ViT encoder pretrained with large-scale MAE self-supervision retrieves clinically similar 3D DWI/ADC volumes in the image domain, and their paired clinician-authored reports ground LLaMA3-8B-Instruct report generation, avoiding explicit image-text alignment; on a 1,831-case multi-institutional internal cohort, a 580-case privacy-preserving external cohort, and the 206-case public ISLES benchmark, PIRTA achieves strong image-domain retrieval (internal mAP@1 of 94.0%) and consistently improves ischemic-territory accuracy, a clinically grounded surrogate for factuality, over direct image-to-text 2D baselines, leading the strongest 2D baseline by +57.2 / +34.5 / +30.6 multi-class accuracy points.

This study developed CARDIAC-FM, a multimodal foundation model that integrates 12-lead ECG and cardiac MRI through self-supervised representation learning and cross-modal contrastive alignment, pretrained on 57,609 paired samples from UK Biobank, improving prediction of incident atrial fibrillation and heart failure, generalizing zero-shot to the Cardiovascular Health Study and the Multi-Ethnic Study of Atherosclerosis, and transferring to cardiac MRI phenotype prediction, time-to-event modelling, and additional outcomes including myocardial infarction, ischaemic stroke, cardiovascular death, and all-cause mortality.

This study developed CARDIAC-FM, a multimodal foundation model that integrates 12-lead ECG and cardiac MRI through self-supervised representation learning and cross-modal contrastive alignment, pretrained on 57,609 paired samples from UK Biobank, improving prediction of incident atrial fibrillation and heart failure, generalizing zero-shot to the Cardiovascular Health Study and the Multi-Ethnic Study of Atherosclerosis, and transferring to cardiac MRI phenotype prediction, time-to-event modelling, and additional outcomes including myocardial infarction, ischaemic stroke, cardiovascular death, and all-cause mortality.