Volet technique et scientifique
Improving Clinician Performance in Classifying EEG Patterns on the Ictal–Interictal Injury Continuum Using Interpretable Machine Learning - 23 mai 2024
An interpretable deep-learning model was developed to classify six types of EEG patterns, trained on over 50,000 samples from ICU patients. Its accuracy was confirmed by medical professionals, comparing AI-assisted and unassisted diagnoses, and validated through statistical measures. Additionally, the model supports the continuum hypothesis of seizure-like brain activity, as shown by data-driven neural network analysis.
Gautam Kamath, analyse les derniers développements en matière de robustesse et de protection de la vie privée. - 28 mai 2024
En se concentrant sur ce problème, il discute de certains développements récents en matière de robustesse et de confidentialité, en soulignant les liens conceptuels et techniques surprenants entre les deux.
Les inconnues connues : Geoff Pleiss, chercheur chez Vector, se penche sur l'incertitude pour rendre les modèles de ML plus précis. - 3 mai 2024
Geoff Pleiss, a member of the Vector Institute, emphasizes the crucial role of uncertainty data in critical safety applications and decision-making processes. The challenge with contemporary neural networks is quantifying uncertainty, given their vast scale and intricate complexity, coupled with a limited grasp of their functionality. Pleiss's work on "deep ensembling," training diverse neural networks to produce variant models, has yielded unexpected results
Promising algorithms to perilous applications: a systematic review of risk stratification tools for predicting healthcare utilisation - 19 juin 2024
Risk stratification tools that predict healthcare utilisation are extensively integrated into primary care systems worldwide, forming a key component of anticipatory care pathways, where high-risk individuals are targeted by preventative interventions. Existing work broadly focuses on comparing model performance in retrospective cohorts with little attention paid to efficacy in reducing morbidity when deployed in different global contexts. We review the evidence supporting the use of such tools in real-world settings, from retrospective dataset performance to pathway evaluation.
Industry- and AI-focused cloud transformation - MIT Technology Review Insights - 24 mai 2024
A successful cloud-based business transformation requires solutions with industry-specific best practices and AI readiness built in.
KAN: Kolmogorov–Arnold Networks - 16 juin 2024
They propose Kolmogorov Arnold Networks (KANs) as promising alternatives to Multi-Layer Perceptrons (MLPs).While MLPs have fixed activation functions on nodes (“neurons”), KANs have learnable activation functions on edges (“weights”). KANs have no linear weights at all – every weight parameter is replaced by a univariate function parametrized as a spline. We show that this seemingly simple change makes KANs outperform MLPs in terms of accuracy and interpretability, on small-scale AI + Science tasks. For accuracy, smaller KANs can achieve comparable or better accuracy than larger MLPs in function fitting tasks. Proposé par David Munger, conseiller-cadre au CEIAVD