Publications

A Roadmap for MEG Foundation Models
Yorguin Mantilla-Ramos
Fouad Lbakali
Catherine Duclos
Annalisa Pascarella
Vanessa Hadid
Oiwi Parker Jones
Foundation models are beginning to reshape brain-signal analysis by moving the field beyond task-specific decoding pipelines toward reusable… (see more) models pretrained on broad neural datasets. Magnetoencephalography (MEG) is a compelling but still underdeveloped target for this shift: it captures human cortical dynamics at millisecond resolution while offering stronger spatial interpretability than EEG, making it especially valuable for source-resolved studies of perception, language, cognition, and clinical brain function. Yet MEG foundation models remain at an early stage, with only a small number of MEG-specific and MEG-inclusive multi-modal models, modest pretraining corpora, and emerging but still limited benchmarks. This perspective lays down the basic concepts needed to understand MEG foundation models and provides a didactic overview of the field's key design choices, including tokenization, sensor- versus source-space representations, sensor-geometry encoding, backbone architectures, self-supervised objectives, and pretraining data. We then offer a roadmap for future development, organized around native MEG pretraining, adaptation of EEG foundation models, transfer from generic time-series models, and multi-modal integration with EEG, fMRI, MRI, behaviour, and stimulus features. We highlight the need for coordinated infrastructure, including diverse and reusable MEG datasets, rigorous evaluation across subjects, sites, tasks, and clinical settings, and responsible data-sharing practices that address consent, privacy, access, and governance.
The 2026 PNPL Competition: Word Classification and Efficient Cross-Subject Generalisation in LibriBrain100
Francesco Mantegna
Gereon Elvers
Dulhan Jayalath
Gilad Landau
Tasha Kim
Miran Ozdogan
Luisa Kurth
Teyun Kwon
SungJun Cho
Benjamin Ballyk
Alex Fung
Anna Greer
Pratik Somaiya
Christian Herff
Greg Farquhar
Brendan Shillingford
Mark Woolrich … (see 1 more)
Oiwi Parker Jones
The ambition of the 2025 PNPL competition (Landau et al., 2025) was to launch a multi-year curriculum for non-invasive speech decoding. Desi… (see more)gned to progress from foundational tasks toward the linguistic complexity required for a practical brain-computer interface (BCI), it set the stage with speech detection and phoneme classification tasks. Winning submissions reached F1-macro scores of 95.6% and 73.6% on the respective tasks (Elvers et al., 2026), highly significant advances. This success was built on the LibriBrain dataset (Özdogan et al., 2025), the largest within-subject MEG dataset recorded at the time with
The 2026 PNPL Competition: Word Classification and Efficient Cross-Subject Generalisation in LibriBrain100
Francesco Mantegna
Gereon Elvers
Dulhan Jayalath
Gilad Landau
Tasha Kim
Miran Ozdogan
Luisa Kurth
Teyun Kwon
SungJun Cho
Benjamin Ballyk
Alex Fung
Anna Greer
Pratik Somaiya
Christian Herff
Greg Farquhar
Brendan Shillingford
Mark Woolrich … (see 1 more)
Oiwi Parker Jones
The ambition of the 2025 PNPL competition (Landau et al., 2025) was to launch a multi-year curriculum for non-invasive speech decoding. Desi… (see more)gned to progress from foundational tasks toward the linguistic complexity required for a practical brain-computer interface (BCI), it set the stage with speech detection and phoneme classification tasks. Winning submissions reached F1-macro scores of 95.6% and 73.6% on the respective tasks (Elvers et al., 2026), highly significant advances. This success was built on the LibriBrain dataset (Özdogan et al., 2025), the largest within-subject MEG dataset recorded at the time with
Complexity-driven feature selection for enhancing tuberculosis detection
Sana Ben Mahjouba
Johannes C. Ayena
Youssef Ouakrim
Simon Grandjean Lapierre
Mihaja Raberahona
Neila Mezghani
Most existing machine learning approaches for tuberculosis (TB) screening typically utilize large, high-dimensional acoustic feature sets wi… (see more)thout examining their intrinsic discriminative power. To address this gap, we introduce a complexity-based feature selection approach that evaluates temporal and spectral descriptors using Fisher score (F1), class-distribution overlap (F2), and Shannon entropy (F4). Applied to the CODA-TB dataset (9772 audio recordings from 1105 participants), the proposed method identified 7 highly informative features from the original 26 features, primarily consisting of mel-frequency cepstral coefficients (MFCC) derivatives and spectral-shape measures. The resulting model achieved performance comparable to full-feature baselines while reducing feature dimensionality by 73% and computational cost by up to 14×. Comparative evaluation against four established feature selection techniques, supported by ablation and statistical analyses, confirmed the efficiency and robustness of the complexity-driven strategy, with no statistically significant loss in performance. These findings highlight the potential of lightweight, interpretable, and computationally efficient models for TB cough-based screening in resource-constrained environments.
Robust inference and correlates from genetic associations with personality
Ted Schwaba
Margaret L. Clapp Sullivan
Wonuola A. Akingbuwa
Kerli Ilves
Peter T. Tanksley
Camille M. Williams
Yavor Dragostinov
Travis T. Mallard
Justin D. Tubbs
Wangjingyi Liao
Lindsay S. Ackerman
Josephine C. M. Fealy
Gibran Hemani
Javier de la Fuente
George Davey Smith
Priya Gupta
Murray B. Stein
Joel Gelernter
Daniel F. Levey
Urmo Võsa … (see 121 more)
Liisi Ausmees
Anu Realo
Tõnu Esko
Mariliis Vaht
Jüri Allik
Tõnu Esko
René Mõttus
Uku Vainik
Gudrun A. Jonsdottir
Gudmar Thorleifsson
Árni Freyr Gunnarsson
Gyda Bjornsdottir
Thorgeir E. Thorgeirsson
Hreinn Stefansson
Kari Stefansson
Rosa Cheesman
Qi Qin
Elizabeth C. Corfield
Helga Ask
Fartein Ask Torvik
Eivind Ystrom
Martin Tesli
Dorret I. Boomsma
Eco J. C. de Geus
Jouke-Jan Hottenga
Dener Cardoso Melo
Harold Snieder
Catharina A. Hartman
Charley Xia
Archie Campbell
Michelle Luciano
Ian J. Deary
W. David Hill
Seon-Kyeong Jang
Scott I. Vrieze
Gonçalo Abecasis
Michelle K. Lupton
Brittany L. Mitchell
Petra V. Viher
Lucía Colodro-Conde
Nicholas G. Martin
Sarah E. Medland
Eske M. Derks
Briar Wormington
Jaakko Kaprio
Karri Silventoinen
Teemu Palviainen
Agnieszka Musial
Kaili Rimfeld
Robert Plomin
Margherita Malanchini
Danielle M. Dick
Fazil Aliev
COGA Collaborators
The Spit for Science Working Group
Laura W. Wesseldijk
Fredrik Ullén
Miriam A. Mosing
Henry R. Kranzler
Yaira Nunez
Sarah Beck
Renato Polimanti
Tobias Edwards
Alexandros Giannelis
Emily A. Willoughby
James J. Lee
Matt McGue
Antonio Terracciano
Michele Marongiu
Edoardo Fiorillo
Francesco Cucca
Angelina R. Sutin
Peter J. van der Most
Albertine J. Oldehinkel
Tina Kretschmer
Andrey A. Shabalin
Anna R. Docherty
Robert F. Krueger
Colin D. Freilich
Binisha H. Mishra
Terho Lehtimäki
Olli T. Raitakari
Mika Kähönen
Aino Saarinen
Henrik Dobewall
Liisa Keltikangas-Järvinen
Klaus Berger
Marisol Herrera-Rivero
Fabian Streit
Swapnil Awasthi
Stephanie H. Witt
Johanna Tuhkanen
Katri Räikkönen
Johan G. Eriksson
Jari Lahti
Gail Davies
Paul Redmond
Adele Taylor
Janie Corley
Tom C. Russ
Marina Ciullo
Teresa Nutile
Yong Qian
Toshiko Tanaka
Luigi Ferrucci
Lea Zillich
Lea Sirignano
K. Paige Harden
Erhan Genç
Patrick D. Gajewski
Stephan Getzmann
Christoph Fraenz
Javier E. Schneider Peñate
Stefanie Lis
Alisha S. M. Hall
Christian Schmahl
Sabine C. Herpertz
Abdel Abdellaoui
Michel G. Nivard
Elliot M. Tucker-Drob
Personality traits describe stable differences in how people think, feel and behave, and how they interact with and experience their social … (see more)and physical environments1,2. Many questions remain unanswered about associations between DNA and personality traits, such as their robustness, their generalizability and the biological and social pathways through which they act. Here we meta-analyse data across 46 cohorts comprising 611,037 to 1.14 million participants with European-like and African-like genomes for genome-wide association studies (GWAS) of the Big Five personality traits (extraversion, agreeableness, conscientiousness, neuroticism and openness to experience), and data from up to 50,725 participants for within-family GWAS. We identify 1,260 lead genetic variants associated with personality, including 824 novel variants3. Common genetic variants explain a moderate 4.8-9.3% of the variance in measures of each trait, and 9.3-13.3% among instruments with typical measurement reliability. Genetic associations with personality are highly consistent but not identical across geography, reporter (self versus close other), age group and measurement instrument, and we find minimal spousal assortment for personality in recent history. In contrast to many other social and behavioural traits4,5, within-family GWAS and polygenic index analyses indicate that genetic associations with personality are minimally confounded by the shared family environment. Polygenic prediction, genetic correlation and Mendelian randomization analyses indicate that personality traits have widespread, potentially causal associations with consequential behaviours and life outcomes. Overall, we find that the genetic architecture of personality is robustly generalizable, minimally confounded and widely relevant to human experience.
Source-Free Class Relearning: Diagnosing Forgetting in Class Unlearning
Zahra Dehghani
Mohammadhadi Shateri
Class unlearning aims to remove a model's ability to recognize designated forget classes while preserving performance on retain classes. How… (see more)ever, low forget accuracy after unlearning does not necessarily mean the class structure has been erased. Approximate unlearning methods can alter classifier decision boundaries while leaving recoverable structure in the representation. Prior work has shown that forget classes can be recovered, but existing approaches require real forget or retain samples, auxiliary data, or reference checkpoints. We study class relearning in a strictly source-free setting, asking whether a forget class can be recovered through a classifier-head update using only the unlearned model. Our approach rests on a theoretical analysis establishing a sufficient alignment condition under which a single gradient step on a synthetic probe set increases the expected logit margin of the forget class. Building on this, we propose a white-box Source-Free Relearning Audit (SFRA), which generates candidate embeddings in representation space and uses model-guided confidence filtering to construct high-confidence retain probes and low-confidence boundary-adjacent probes that are relabelled as the forget class. Gaussian sampling and Softmax confidence are used by default, while ablations with alternative proposal distributions and uncertainty criteria show that recoverability is not specific to these choices. To quantify recoverability, we introduce the Relearning Score (RS), which jointly measures forget-class recovery and retain-accuracy preservation, and report class-matched
AN AGENTIC FRAMEWORK FOR INTERROGATABLE AND INTERPRETABLE SURVIVAL ANALYSIS
Juan Felipe Duran
Shirin Abbasinejad Enger
Android Automotive OS under the hood: Architecture and security threat classification
Faiyaz Khan
Hanbo Yu
Junjie Wu
Natalia Stakhanova
Benjamin C.M. Fung
Beyond Hard Writes and Rigid Preservation: Soft Recursive Least-Squares for Lifelong LLM Editing
Xinyu Wang
Yu Gu
Peng Lu
Yufei Cui
Xiao-Wen Chang
Model editing updates a pre-trained LLM with new facts or rules without retraining while preserving unrelated behavior. In real deployment, … (see more)edits arrive as long streams, creating a plasticity-stability dilemma: repeated locate-then-edit "hard writes" can accumulate interference over time, while rigid preservation constraints may protect only explicitly constrained directions, allowing past edits or unconstrained behaviors to deviate. We propose RLSEdit, a recursive least-squares editor for long sequential editing. RLSEdit formulates editing as an online quadratic optimization with soft constraints, minimizing a cumulative key-value fitting objective together with two regularizers that control deviation from the pre-trained weights and from a designated anchor mapping. This objective admits an efficient Woodbury-based online recursion, with per-edit cost independent of history length and scaling only with the current edit size. We further provide deviation bounds and an asymptotic characterization of the adherence-preservation trade-off in the many-edits regime. Experiments on CounterFact and ZsRE across multiple model families show stable scaling to 10K edits, outperforming strong baselines in both edit success and holistic stability, while retaining early edits and preserving general capabilities on GLUE and held-out reasoning/code benchmarks.
CRAG-MM-Diagnostics: Enabling Stage-Wise Analysis of Knowledge-Intensive VQA
Hanseok Oh
Hyunji Lee
Paul Liang
Knowledge-Intensive Visual Question Answering (KI-VQA) benchmarks evaluate Vision-Language Models (VLMs) as multimodal knowledge assistants … (see more)by requiring external information beyond a provided image to answer questions. KI-VQA involves multiple sub-problems -referring expression understanding, visual grounding, object recognition, knowledge retrieval, and reasoning-yet existing benchmarks typically report only end-task accuracy, obscuring where failures arise. To analyze the full KI-VQA pipeline, we introduce CRAG-MM-Diagnostics, a diagnostic benchmark with stage-wise data annotations that isolate 1) language-based visual grounding, 2) object identification, and 3) knowledge retrieval and reasoning. We evaluate fully parametric and retrieval-augmented VLMs, providing fine-grained analyses using newly collected metadata, such as target ROIs, entity names, and visual complexity scores. Our results point to knowledge retrieval and reasoning as the primary bottleneck, but also highlight issues in the other parts of the KI-VQA pipeline, such as the fact that VLMs struggle with target object identification or that image retrievers struggle to integrate textual cues. These findings expose fundamental limitations in current KI-VQA systems and motivate stage-aware evaluation. We, lastly, leverage these findings to propose a grounded bimodal RAG pipeline that integrates a visual grounding module to crop targets before image retrieval, boosting GPT-5 and Qwen's respective accuracies by 13.3 and 8.5 percentage points.
Creating reference data for early hospital outbreak detection algorithms from experts’ ratings
Brice Leclère
Didier Lepelletier
David L. Buckeridge
BACKGROUND Early outbreak detection algorithms can be useful tools to prevent healthcare-acquired infections in hospitals. However, their de… (see more)velopment and evaluation are hindered by the lack of available labelled data. AIM The aim of this study was to build a reference dataset for outbreak detection in hospitals using two different consensus approaches used to build this dataset. METHODS 25 Canadian and French experts were asked to review one-year time series of weekly incidence from different types of microorganisms, based on the data of a French university hospital. For each time series, experts also add access to additional surveillance data (locations and investigations). Each time series was submitted to three experts, whose role was to identify potential outbreak periods and rank their probability on a web platform. These rankings were summarized using two approaches: a majority vote wherein the most prevalent ranking was used for each week, and a hidden Markov model (HMM) in which the answers of the experts were used as observable variables that related to latent epidemiological states. FINDINGS three experts reviewed a total of 36 times series i.e., 1899 surveillance weeks for 14 different types of microorganisms. Overall, the concordance between the two approaches was the highest for identifying high-ranking weeks. All rankings considered, 89 potential events were identified by majority vote and 96 by the HMM. CONCLUSION We constructed a reliable reference standard data set for the development, evaluation and comparison of algorithms for nosocomial outbreak detection within hospitals.
Exponentiated gradient learning yields brain-like synaptic distributions
Jonathan H. Cornford
Kaiwen Sheng
Brendan A. Bicknell
Beverley A. Clark
Blake A. Richards
Artificial neural networks trained with gradient descent (GD) are widely used to model brain learning, but GD allows synapses to switch betw… (see more)een excitatory and inhibitory and produces weight distributions unlike those observed experimentally. We show that an alternative rule, exponentiated gradient (EG), avoids these problems because its multiplicative, exponential weight updates mathematically prohibit synapses from changing sign and preferentially scale already-large weights, naturally generating log-normal distributions. This mechanism also concentrates task-relevant computation onto a smaller subset of strong connections, which we show makes EG-trained recurrent neural networks more robust to synaptic pruning and better able to learn from sparsely relevant inputs than GD-trained networks, while matching GD's performance on standard cognitive tasks. These findings tie a normative optimization principle directly to observed features of synaptic organization in the brain.