Université Mohammed VI Polytechnique

3 related articles

Subscribe to updates

bioRxiv Nov 6, 2026
Single-cell perturbation prediction stops chasing the average: scDisentangle keeps each cell's identity in its counterfactuals
Single-cell perturbation-response prediction has struggled to beat simple mean baselines, and population-average metrics cannot show whether a prediction keeps the source cell's identity. scDisentangle splits each cell's expression into attribute embeddings and a sequence of residual factors extracted one after another, then generates counterfactuals at inference by replacing only the attribute embeddings while holding cell-specific factors fixed. On interferon stimulation, liver infection and unseen combinatorial genetic perturbations it beat the models and mean baselines compared in each benchmark, and in a myocarditis cohort it predicted donor-specific heart-infiltrating CD8+ T cell states from peripheral blood.
Single-cell perturbation-response prediction has struggled to beat simple mean baselines, and population-average metrics cannot show whether a prediction keeps the source cell's identity. scDisentangle splits each cell's expression into attribute embeddings and a sequence of residual factors extracted one after another, then generates counterfactuals at inference by replacing only the attribute embeddings while holding cell-specific factors fixed. On interferon stimulation, liver infection and unseen combinatorial genetic perturbations it beat the models and mean baselines compared in each benchmark, and in a myocarditis cohort it predicted donor-specific heart-infiltrating CD8+ T cell states from peripheral blood.
Hebrew University of Jerusalem Université Mohammed VI Polytechnique
Salim El Mejjad（First author）Tariq Daouda（Corresponding author）
scDisentangle Single-cell perturbation response prediction Disentangled Representation Learning
researchsquare.com Nov 9, 2026
Heterogeneous graph attention stops at path endpoints no more: TAHAN groups intermediate nodes by type and leads DBLP and IMDB classification means
Attention aggregation on heterogeneous graphs has long treated node types as static labels, and metapath aggregation can overlook the semantic contribution of intermediate nodes. TAHAN instead collects neighbors along metapath prefixes, merges them across metapaths and levels by node type, then fuses them with within-type attention followed by across-type attention. On DBLP and IMDB node classification its mean Macro-F1 and Micro-F1 exceed the listed baselines, with both IMDB gains significant after Holm correction; on ACM it falls below SeHGNN.
Attention aggregation on heterogeneous graphs has long treated node types as static labels, and metapath aggregation can overlook the semantic contribution of intermediate nodes. TAHAN instead collects neighbors along metapath prefixes, merges them across metapaths and levels by node type, then fuses them with within-type attention followed by across-type attention. On DBLP and IMDB node classification its mean Macro-F1 and Micro-F1 exceed the listed baselines, with both IMDB gains significant after Holm correction; on ACM it falls below SeHGNN.
Université Mohammed VI Polytechnique Sidi Mohamed Ben Abdellah University Cadi Ayyad University
Amine Lakhder（First author, Corresponding author）
TAHAN Heterogeneous Graph Neural Networks Type-Aware Attention Node Classification Link Prediction
arXiv Oct 2, 2026
TACO optimizer cuts LLM fine-tuning optimizer-state memory 174x and fine-tunes 30-32B models on a single 80 GB H100
The authors propose TACO, an optimizer that computes the exact steepest-descent direction under a dimension-normalized 1-to-1 operator norm by taking the sign of the largest-magnitude entry in each column of two-dimensional weight matrices, keeping only a small set of low-precision gradient components per column; on OPT-13B it reduces persistent optimizer state 174x versus AdamW8bit (27.7 GB to 0.16 GB) and peak training memory 2.9x (80.6 GB to 27.5 GB) with comparable accuracy and runtime, and it enables full-parameter fine-tuning of 30-32B-parameter models on a single 80 GB H100.
The authors propose TACO, an optimizer that computes the exact steepest-descent direction under a dimension-normalized 1-to-1 operator norm by taking the sign of the largest-magnitude entry in each column of two-dimensional weight matrices, keeping only a small set of low-precision gradient components per column; on OPT-13B it reduces persistent optimizer state 174x versus AdamW8bit (27.7 GB to 0.16 GB) and peak training memory 2.9x (80.6 GB to 27.5 GB) with comparable accuracy and runtime, and it enables full-parameter fine-tuning of 30-32B-parameter models on a single 80 GB H100.
University of Central Florida Université Mohammed VI Polytechnique
Jichao Jiang（First author）
Ternary Absolute-max Column-wise One-sparse Optimizer Fine-tuning Large language models

You’re all caught up

Single-cell perturbation prediction stops chasing the average: scDisentangle keeps each cell's identity in its counterfactuals

Heterogeneous graph attention stops at path endpoints no more: TAHAN groups intermediate nodes by type and leads DBLP and IMDB classification means

TACO optimizer cuts LLM fine-tuning optimizer-state memory 174x and fine-tunes 30-32B models on a single 80 GB H100