Publications

Empowering Clinicians with Medical Decision Transformers: A Framework for Sepsis Treatment
Aamer Abdul Rahman
Pranav Agarwal
Rita Noumeir
Philippe Jouvet
Vincent Michalski
S Ebrahimi Kahou
Offline reinforcement learning has shown promise for solving tasks in safety-critical settings, such as clinical decision support. Its appli… (voir plus)cation, however, has been limited by the lack of interpretability and interactivity for clinicians. To address these challenges, we propose the medical decision transformer (MeDT), a novel and versatile framework based on the goal-conditioned reinforcement learning paradigm for sepsis treatment recommendation. MeDT uses the decision transformer architecture to learn a policy for drug dosage recommendation. During offline training, MeDT utilizes collected treatment trajectories to predict administered treatments for each time step, incorporating known treatment outcomes, target acuity scores, past treatment decisions, and current and past medical states. This analysis enables MeDT to capture complex dependencies among a patient's medical history, treatment decisions, outcomes, and short-term effects on stability. Our proposed conditioning uses acuity scores to address sparse reward issues and to facilitate clinician-model interactions, enhancing decision-making. Following training, MeDT can generate tailored treatment recommendations by conditioning on the desired positive outcome (survival) and user-specified short-term stability improvements. We carry out rigorous experiments on data from the MIMIC-III dataset and use off-policy evaluation to demonstrate that MeDT recommends interventions that outperform or are competitive with existing offline reinforcement learning methods while enabling a more interpretable, personalized and clinician-directed approach.
2024-07-27
ArXiv (prépublication)
doi.org
arxiv.org
On the benefits of pixel-based hierarchical policies for task generalization
T. Cristea-Platon
Bogdan Mazoure
Josh Susskind
Walter Talbott
Reinforcement learning practitioners often avoid hierarchical policies, especially in image-based observation spaces. Typically, the single-… (voir plus)task performance improvement over flat-policy counterparts does not justify the additional complexity associated with implementing a hierarchy. However, by introducing multiple decision-making levels, hierarchical policies can compose lower-level policies to more effectively generalize between tasks, highlighting the need for multi-task evaluations. We analyze the benefits of hierarchy through simulated multi-task robotic control experiments from pixels. Our results show that hierarchical policies trained with task conditioning can (1) increase performance on training tasks, (2) lead to improved reward and state-space generalizations in similar tasks, and (3) decrease the complexity of fine tuning required to solve novel tasks. Thus, we believe that hierarchical policies should be considered when building reinforcement learning architectures capable of generalizing between tasks.
2024-07-26
ArXiv (prépublication)
doi.org
arxiv.org
Canada's Provincial Covid-19 Pandemic Modelling Efforts: A Review of Mathematical Models and Their Impacts on the Responses
Yiqing Xia
Jorge Luis Flores Anato
Caroline Colijin
Naveed Janjua
Michael Otterstatter
Mike Irvine
Tyler Williamson
Marie B. Varughese
Michael Li
Nathaniel Osgood
David J. D. Earn
Beate Sander
Lauren E. Cipriano
Kumar Murty
Fanyu Xiu
Arnaud Godin
David L Buckeridge
Amy Hurford
Sharmistha Mishra
Mathieu Maheu-Giroux
2024-07-24
Canadian Journal of Public Health (publié)
doi.org
Multi-Fidelity Active Learning with GFlowNets
Alex Hernandez-Garcia
Nikita Saxena
Moksh J. Jain
Cheng-Hao Liu
Yoshua Bengio
In the last decades, the capacity to generate large amounts of data in science and engineering applications has been growing steadily. Meanw… (voir plus)hile, the progress in machine learning has turned it into a suitable tool to process and utilise the available data. Nonetheless, many relevant scientific and engineering problems present challenges where current machine learning methods cannot yet efficiently leverage the available data and resources. For example, in scientific discovery, we are often faced with the problem of exploring very large, high-dimensional spaces, where querying a high fidelity, black-box objective function is very expensive. Progress in machine learning methods that can efficiently tackle such problems would help accelerate currently crucial areas such as drug and materials discovery. In this paper, we propose the use of GFlowNets for multi-fidelity active learning, where multiple approximations of the black-box function are available at lower fidelity and cost. GFlowNets are recently proposed methods for amortised probabilistic inference that have proven efficient for exploring large, high-dimensional spaces and can hence be practical in the multi-fidelity setting too. Here, we describe our algorithm for multi-fidelity active learning with GFlowNets and evaluate its performance in both well-studied synthetic tasks and practically relevant applications of molecular discovery. Our results show that multi-fidelity active learning with GFlowNets can efficiently leverage the availability of multiple oracles with different costs and fidelities to accelerate scientific discovery and engineering design.
2024-07-24
TMLR (accepté)
doi.org
openreview.net
Development of Error Passing Network for Optimizing the Prediction of VO$_2$ peak in Childhood Acute Leukemia Survivors
Nicolas Raymond
Hakima Laribi
Maxime Caru
Mehdi Mitiche
Valerie Marcil
Maja Krajinovic
Daniel Curnier
Daniel Sinnett
Martin Vallières
Approximately two-thirds of survivors of childhood acute lymphoblastic leukemia (ALL) cancer develop late adverse effects post-treatment. Pr… (voir plus)ior studies explored prediction models for personalized follow-up, but none integrated the usage of neural networks to date. In this work, we propose the Error Passing Network (EPN), a graph-based method that leverages relationships between samples to propagate residuals and adjust predictions of any machine learning model. We tested our approach to estimate patients’ \vo peak, a reliable indicator of their cardiac health. We used the EPN in conjunction with several baseline models and observed up to 12.16% improvement in the mean average percentage error compared to the last established equation predicting \vo peak in childhood ALL survivors. Along with this performance improvement, our final model is more efficient considering that it relies only on clinical variables that can be self-reported by patients, therefore removing the previous need of executing a resource-consuming physical test.
2024-07-23
Proceedings of the fifth Conference on Health, Inference, and Learning (publié)
proceedings.mlr.press
proceedings.mlr.press
In value-based deep reinforcement learning, a pruned network is a good network
Johan Obando-Ceron
Aaron Courville
Pablo Samuel Castro
Recent work has shown that deep reinforcement learning agents have difficulty in effectively using their network parameters. We leverage pri… (voir plus)or insights into the advantages of sparse training techniques and demonstrate that gradual magnitude pruning enables value-based agents to maximize parameter effectiveness. This results in networks that yield dramatic performance improvements over traditional networks, using only a small fraction of the full network parameters.
2024-07-22
ICML (Accept (Poster))
doi.org
proceedings.mlr.press
A Waddington landscape for prototype learning in generalized Hopfield networks
Nacer Eddine Boukacem
Allen Leary
Robin Theriault
Felix Gottlieb
Madhav Mani
Paul François
Networks in machine learning offer examples of complex high-dimensional dynamical systems reminiscent of biological systems. Here, we study … (voir plus)the learning dynamics of Generalized Hopfield networks, which permit a visualization of internal memories. These networks have been shown to proceed through a 'feature-to-prototype' transition, as the strength of network nonlinearity is increased, wherein the learned, or terminal, states of internal memories transition from mixed to pure states. Focusing on the prototype learning dynamics of the internal memories we observe a strong resemblance to the canalized, or low-dimensional, dynamics of cells as they differentiate within a Waddingtonian landscape. Dynamically, we demonstrate that learning in a Generalized Hopfield Network proceeds through sequential 'splits' in memory space. Furthermore, order of splitting is interpretable and reproducible. The dynamics between the splits are canalized in the Waddington sense -- robust to variations in detailed aspects of the system. In attempting to make the analogy a rigorous equivalence, we study smaller subsystems that exhibit similar properties to the full system. We combine analytical calculations with numerical simulations to study the dynamical emergence of the feature-to-prototype transition, and the behaviour of splits in the landscape, saddles points, visited during learning. We exhibit regimes where saddles appear and disappear through saddle-node bifurcations, qualitatively changing the distribution of learned memories as the strength of the nonlinearity is varied -- allowing us to systematically investigate the mechanisms that underlie the emergence of Waddingtonian dynamics. Memories can thus differentiate in a predictive and controlled way, revealing new bridges between experimental biology, dynamical systems theory, and machine learning.
2024-07-22
Physical Review Research (publié)
doi.org
arxiv.org
Wasserstein Distributionally Robust Shallow Convex Neural Networks
Julien Pallage
Antoine Lesage-Landry
2024-07-22
ArXiv (prépublication)
doi.org
arxiv.org
A Rapid Method for Impact Analysis of Grid-Edge Technologies on Power Distribution Networks
Feng Li
Ilhan Kocar
Antoine Lesage-Landry
This paper presents a novel rapid estimation method (REM) to perform stochastic impact analysis of grid-edge technologies (GETs) to the powe… (voir plus)r distribution networks. The evolution of network states' probability density functions (PDFs) in terms of GET penetration levels are characterized by the Fokker-Planck equation (FPE). The FPE is numerically solved to compute the PDFs of network states, and a calibration process is also proposed such that the accuracy of the REM is maintained for large-scale distribution networks. The approach is illustrated on a large-scale realistic distribution network using a modified version of the IEEE 8500 feeder, where electric vehicles (EVs) or photovoltaic systems (PVs) are installed at various penetration rates. It is demonstrated from quantitative analyses that the results from our proposed approach have negligible errors comparing with those obtained from Monte Carlo simulations.
2024-07-20
2024 IEEE Power & Energy Society General Meeting (PESGM) (publié)
doi.org
T2VIndexer: A Generative Video Indexer for Efficient Text-Video Retrieval
Yili Li
Jing Yu
Keke Gai
Bang Liu
Gang Xiong
Qi Wu
Current text-video retrieval methods mainly rely on cross-modal matching between queries and videos to calculate their similarity scores, wh… (voir plus)ich are then sorted to obtain retrieval results. This method considers the matching between each candidate video and the query, but it incurs a significant time cost and will increase notably with the increase of candidates. Generative models are common in natural language processing and computer vision, and have been successfully applied in document retrieval, but their application in multimodal retrieval remains unexplored. To enhance retrieval efficiency, in this paper, we introduce a model-based video indexer named T2VIndexer, which is a sequence-to-sequence generative model directly generating video identifiers and retrieving candidate videos with constant time complexity. T2VIndexer aims to reduce retrieval time while maintaining high accuracy. To achieve this goal, we propose video identifier encoding and query-identifier augmentation approaches to represent videos as short sequences while preserving their semantic information. Our method consistently enhances the retrieval efficiency of current state-of-the-art models on four standard datasets. It enables baselines with only 30%-50% of the original retrieval time to achieve better retrieval performance on MSR-VTT (+1.0%), MSVD (+1.8%), ActivityNet (+1.5%), and DiDeMo (+0.2%). The code is available at https://anonymous.4open.science/r/T2VIndexer-40BE.
2024-07-19
acmmm.org/ACMMM/2024/Conference (présentation orale)
openreview.net
openreview.net
ANDES, the high resolution spectrograph for the ELT: science goals, project overview, and future developments
Alessandro Marconi
Artur R. Abreu
Vardan Adibekyan
Valentina Alberti
Simon Albrecht
Jailson Alcaniz
Matteo Aliverti
Carlos Allende Prieto
Julian Alvarado-Gomez
Catarina Alves
Pedro J. Amado
Manuel Amate
Michael Andersen
Simone Antoniucci
E. Artigau
Christophe Bailet
Clark E. Baker
Veronica Baldini
Andrea Balestra
S.A. Barnes … (voir 271 de plus)
Frédérique Baron
Susana Barros
Svend-Marian Bauer
Mathilde Beaulieu
Olga Bellido-Tirado
Björn Benneke
Thomas Bensby
Edwin Bergin
P. Berio
Katia Biazzo
Laurent Bigot
Arjan Bik
Jayne L. Birkby
Nicolas Blind
Olivier Boebion
Isabelle Boisse
Emeline Bolmont
J. S. Bolton
Marco Bonaglia
Xavier Bonfils
Lea Bonhomme
Francesco Borsa
Jean-Claude Bouret
Alexis Brandeker
Wolfgang Brandner
Christopher H. Broeg
Matteo Brogi
Denis Brousseau
Anna Brucalassi
Joar G. Brynnel
Lars A. Buchhave
David F. Buscher
Lorenzo Cabona
A. Cabral
Alexandre Cabral
Giorgio Calderone
Rocío Calvo-Ortega
Faustine Cantalloube
Bruno L. Canto Martins
Luca Carbonaro
Yan Caujolle
Gaël Chauvin
Bruno Chazelas
Anne-Laure L. Cheffot
Yuk Shan Cheng
Andrea Chiavassa
Lise B. Christensen
Roberto Cirami
Michele Cirasuolo
Neil J. Cook
Ryan Cooke
Igor Coretti
Stefano Covino
Nicolas B. Cowan
Giovanni Cresci
Stefano Cristiani
Vanderlei Cunha Parro
Guido Cupani
Valentina D'Odorico
Kamalesh Dadi
Izan C. de Castro Leão
Annalisa De Cia
Jose R. De Medeiros
Florian Debras
Michael Debus
Alain Delorme
Olivier Demangeon
Frederic Derie
M. Dessauges-Zavadsky
Paolo Di Marcantonio
Simona Di Stefano
Frank Dionies
Armando Domiciano de Souza
René Doyon
Jennifer S. Dunn
Sébastien E. Egner
David Ehrenreich
Joao P. Faria
Debora Ferruzzi
Chiara Feruglio
Martin Fisher
Adriano Fontana
B S. Frank
C. Fuesslein
M. Fumagalli
Thierry Fusco
Johan P. U. Fynbo
O. Gabella
W. Gaessler
E. Gallo
X. Gao
L. Genolet
M. Genoni
P. Giacobbe
E. Giro
R. S. Gonçalves
O. A. Gonzalez
J. I. González-Hernández
C. Gouvret
F. Gracia Témich
M. G. Haehnelt
C. Haniff
A. Hatzes
R. Helled
H. J. Hoeijmakers
I. Hughes
Philipp Huke
Y. Ivanisenko
A. S. Järvinen
S. P. Järvinen
A. Kaminski
J. Kern
J. Knoche
A. Kordt
H. Korhonen
A. Korn
D. Kouach
G. Kowzan
L. Kreidberg
M. Landoni
A. A. Lanotte
A. Lavail
B. Lavie
D. Lee
M. Lehmitz
Jian Li
Wei Li
J. Liske
C. Lovis
S. Lucatello
D. Lunney
M. J. MacIntosh
N. Madhusudhan
L. Magrini
R. Maiolino
J. Maldonado
L. Malo
A. W. S. Man
T. Marquart
C. M. J. Marques
E. L. Marques
P. Martinez
A. M. Martins
C. J. A. P. Martins
J. H. C. Martins
P. Maslowski
C. Mason
E. Mason
R. A. McCracken
M. A. F. Melo e Sousa
P. Mergo
G. Micela
D. Milaković
P. Mollière
M. A. Monteiro
D. Montgomery
C. Mordasini
J. Morin
A. Mucciarelli
M. T. Murphy
M. N'Diaye
N. Nardetto
B. Neichel
N. Neri
A. T. Niedzielski
E. Niemczura
B. Nisini
L. Nortmann
P. Noterdaeme
N. J. Nunes
L. Oggioni
F. Olchewsky
E. Oliva
H. Önel
L. Origlia
G. Östlin
N. N.-Q. Ouellette
Enric Pallé
P. Papaderos
G. Pariani
L. Pasquini
J. Peñate Castro
F. Pepe
C. Peroux
L. Perreault Levasseur
Sandrine Perruchot
P. Petit
Oliver Pfuhl
L. Pino
Javier Piqueras
N. Piskunov
A. Pollo
K. Poppenhaeger
M. Porru
J. Puschnig
A. Quirrenbach
Emily Rauscher
R. Rebolo
E. M. A. Redaelli
S. Reffert
D. T. Reid
A. Reiners
P. Richter
M. Riva
S. Rivoire
C. Rodríguez-López
I. U. Roederer
D. Romano
M. Roth
S. Rousseau
J. Rowe
A. Saccardi
S. Salvadori
N. Sanna
N. C. Santos
P. Santos Diaz
Jorge Sanz-Forcada
M. Sarajlic
J.-F. Sauvage
D. Savio
A. Scaudo
S. Schäfer
R. P. Schiavon
T. M. Schmidt
C. Selmi
R. Simoes
A. Simonnin
S. Sivanandam
M. Sordet
R. Sordo
F. Sortino
D. Sosnowska
S. G. Sousa
A. Spang
R. Spiga
E. Stempels
J. R. Y. Stevenson
Klaus G. Strassmeier
A. Suárez Mascareño
A. Sulich
X. Sun
N. R. Tanvir
F. Tenegi-Sanginés
S. Thibault
S. J. Thompson
P. Tisserand
A. Tozzi
M. Turbet
J.-P. Véran
Julien Veran
P. Vallée
I. Vanni
R. Varas
A. Vega-Moreno
K. A. Venn
A. Verma
J. Vernet
M. Viel
G. Wade
C. Waring
M. Weber
J. Weder
B. Wehbé
J. Weingrill
M. Woche
M. Xompero
E. Zackrisson
A. Zanutta
M. R. Zapatero Osorio
M. Zechmeister
J. Zimara
2024-07-17
Ground-based and Airborne Instrumentation for Astronomy X (publié)
doi.org
arxiv.org
Curriculum Frameworks and Educational Programs in AI for Medical Students, Residents, and Practicing Physicians: Scoping Review (Preprint)
Raymond Tolentino
Ashkan Baradaran
Genevieve Gore
Pierre Pluye
Samira Abbasgholizadeh-Rahimi
BACKGROUND

The successful integration of artificial intelligence (AI) in… (voir plus)to clinical practice is contingent upon physicians’ comprehension of AI principles and its applications. Therefore, it is essential for medical education curricula to incorporate AI topics and concepts, providing future physicians with the foundational knowledge and skills needed. However, there is a knowledge gap in the current understanding and availability of structured AI curriculum frameworks tailored for medical education, which serve as vital guides for instructing and facilitating the learning process.

 OBJECTIVE

The overall aim of this study is to synthesize knowledge from the literature on curriculum frameworks and current educational programs that focus on the teaching and learning of AI for medical students, residents, and practicing physicians.

 METHODS

We followed a validated framework and the Joanna Briggs Institute methodological guidance for scoping reviews. An information specialist performed a comprehensive search from 2000 to May 2023 in the following bibliographic databases: MEDLINE (Ovid), Embase (Ovid), CENTRAL (Cochrane Library), CINAHL (EBSCOhost), and Scopus as well as the gray literature. Papers were limited to English and French languages. This review included papers that describe curriculum frameworks for teaching and learning AI in medicine, irrespective of country. All types of papers and study designs were included, except conference abstracts and protocols. Two reviewers independently screened the titles and abstracts, read the full texts, and extracted data using a validated data extraction form. Disagreements were resolved by consensus, and if this was not possible, the opinion of a third reviewer was sought. We adhered to the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) checklist for reporting the results.

 RESULTS

Of the 5104 papers screened, 21 papers relevant to our eligibility criteria were identified. In total, 90% (19/21) of the papers altogether described 30 current or previously offered educational programs, and 10% (2/21) of the papers described elements of a curriculum framework. One framework describes a general approach to integrating AI curricula throughout the medical learning continuum and another describes a core curriculum for AI in ophthalmology. No papers described a theory, pedagogy, or framework that guided the educational programs.

 CONCLUSIONS

This review synthesizes recent advancements in AI curriculum frameworks and educational programs within the domain of medical education. To build on this foundation, future researchers are encouraged to engage in a multidisciplinary approach to curriculum redesign. In addition, it is encouraged to initiate dialogues on the integration of AI into medical curriculum planning and to investigate the development, deployment, and appraisal of these innovative educational programs.

 INTERNATIONAL REGISTERED REPORT
2024-07-17
JMIR Medical Education (publié)
doi.org