#multitask-learning News & Analysis

3 articles tagged with #multitask-learning. AI-curated summaries with sentiment analysis and key takeaways from 50+ sources.

3 articles

AINeutralarXiv – CS AI · Mar 267/10

🧠

Entire Space Counterfactual Learning for Reliable Content Recommendations

Researchers developed ESCM² (Entire Space Counterfactual Multitask Model), a new framework that improves post-click conversion rate estimation in recommender systems by addressing intrinsic estimation bias and false independence assumptions. The model-agnostic approach incorporates counterfactual learning to enhance recommendation accuracy and has been validated on large-scale industrial datasets.

AIBullisharXiv – CS AI · Mar 176/10

🧠

Resolving Interference (RI): Disentangling Models for Improved Model Merging

Researchers have developed Resolving Interference (RI), a new framework that improves AI model merging by reducing cross-task interference when combining specialized models. The method makes models functionally orthogonal to other tasks using only unlabeled data, improving merging performance by up to 3.8% and generalization by up to 2.3%.

AINeutralarXiv – CS AI · Mar 54/10

🧠

Directional Neural Collapse Explains Few-Shot Transfer in Self-Supervised Learning

Researchers propose directional CDNV (decision-axis variance) as a key geometric quantity explaining why self-supervised learning representations transfer well with few labels. The study shows that small variability along class-separating directions enables strong few-shot transfer and low interference across multiple tasks.