Publications

SSD: Spatially Speculative Decoding Accelerates Autoregressive Image Generation
Shilong Xiang
Zirui Zhang
Lijun Yu
Chengzhi Mao
Autoregressive image models treat images as 1D token sequences, inheriting the next-token factorization of language models. This flattening … (voir plus)discards a useful property of images: nearby tokens are correlated in two dimensions, not one. We introduce Spatially Speculative Decoding (SSD), an inference-time decoding framework that exploits this spatial structure. Rather than speculating only along the flattened sequence, SSD predicts both the adjacent horizontal token and the token directly below it, allowing multiple spatial directions to advance in parallel. This reduces the number of backbone forward evaluations and alleviates the memory bottleneck of autoregressive decoding. SSD accelerates image generation by up to 11.03x in wall-clock time while maintaining generation quality on DPG-Bench and GenEval. These results show that spatial structure provides a simple and effective source of parallelism for autoregressive image generation.
2026-06-17
arXiv (prépublication)
doi.org
arxiv.org
Timely Availability and Accessibility of Health Data: Meeting the Challenge of Pan-Canadian Health Charter Principle 6
Kimberlyn McGrail
David L. Buckeridge
Pan-Canadian Health Data Charter Principle 6 envisions health data that are timely, accessible, meaningful, and comprehensive. This paper ex… (voir plus)amines three dimensions of this principle: what is envisioned for health data, for whom, and for what purposes. Comprehensive data include publicly funded care, privately paid services, patient-reported outcomes, and social determinants. To be meaningful, data require standardization and quality assurance. Timely access ranges from real-time clinical use to dependable research access. Accessibility depends on interoperable systems and secure environments. Current practices fall short across all dimensions, with fragmented, incomplete, and delayed data limiting timely access and effective use. Access and coverage remain uneven, reflecting structural and policy barriers. Despite these challenges, there are promising initiatives such as efforts to enhance national data stewardship. Advancing Principle 6 will require involvement of all interests in health data, including data stewards, policy-makers, providers, data users, and members of the public.
2026-06-17
Healthcare Management Forum (publié)
doi.org
A call to integrate animal movement into biodiversity indicators
Ruth Y. Oliver
Katherine Hébert
Lacey F. Hughey
Luca Börger
Francesca Cagnacci
Nathan W. Cooper
Sarah C. Davidson
Andrew Gonzalez
Autumn‐Lynn Harrison
Jessica M. Kendall-Bar
Katie L. Millette
Joanna Mills Flemming
Thomas Mueller
Will Rogers
Talia Speaker
Jared A. Stabach
Marlee A. Tucker
Wenjing Xu
Scott W. Yanco
Briana Abrahms … (voir 35 de plus)
Sara Beery
Roxanne S. Beltran
Lily K. Bentley
Larissa T. Beumer
Mary E. Bowers
Steven W. J. Canty
Ying‐Chi Chan
Juliet Cohen
Grant M. Connette
Eduardo Cuevas
Tammy E. Davies
Daniel C. Dunn
Diego Ellis‐Soto
Antonio Ferraz
John Fieberg
Kimberly R. Hall
Neil Hammerschlag
Anne G. Hertel
Dongmin Kim
Samara Manzin
Clive R. McMahon
Robin Naidoo
Aidin Niamir
A. Justin Nowakowski
Matthew B. Ogburn
Jonathan Pye
José Manuel Reyes‐González
Nicholas J. Russo
Christian Rutz
Amy L. Scarpignato
Stella F. Uiterwaal
Raqib Valli
Alessandra Vidal Meza
George Wittemyer
Laura J. Pollock
2026-06-16
Nature Reviews Biodiversity (publié)
doi.org
Can In-Context Learning Support Intrinsic Curiosity?
Eric Elmoznino
Sangnie Bhardwaj
Johannes Von Oswald
Rajai Nasser
Blaise Agüera y Arcas
João Sacramento
Rif A. Saurous
Guillaume Lajoie
Effective machine learning depends not only on how we model data, but also on what data we choose to collect. While large sequence models ha… (voir plus)ve revolutionized data modeling, the problem of automated data selection, or"intrinsic curiosity", remains a significant challenge. Classic approaches incentivize exploration by rewarding an agent based on its"learning progress", which measures how much a newly acquired observation improves a world model's predictive ability. However, evaluating these rewards traditionally requires expensive inner loops of gradient descent updates within each trajectory, rendering them computationally impractical at scale. In this work, we investigate whether the emergent in-context learning (ICL) capabilities of sequence models can eliminate this bottleneck by serving as immediate, update-free world models. Specifically, we evaluate whether an exploration policy can be trained to maximize learning progress, using solely the prediction errors and counterfactual context manipulations of an in-context learner. We first prove that in general Markov decision processes, this is in fact impossible in an unbiased way: the resulting intrinsic rewards either suffer from nuisance terms that bias their estimation of true learning progress, or they cannot be implemented using an in-context learner's prediction errors. Conversely, we prove a positive result for a broad subclass of non-temporal settings, encompassing active learning and Bayesian Experimental Design: here, ICL-derived rewards successfully bound and asymptotically converge to the true learning progress. We corroborate our theory with controlled experiments across continuous and symbolic environments, demonstrating that our ICL-driven framework successfully trains curious data-collection policies that explore optimally.
2026-06-16
arXiv (prépublication)
doi.org
arxiv.org
Convex training of Lipschitz-regularized shallow neural networks
Chao Yin
Antoine Lesage-Landry
In this work, we introduce a training procedure for shallow neural networks that promotes robustness against adversarial attacks. We solve a… (voir plus) non-convex Lipschitz-regularized training program by introducing a convex restriction that can be efficiently solved to global optimality. Our approach can be employed as a post-processing step by taking a pre-trained network as an initial solution to then solving the convex program whose optimal network is guaranteed to be no worse than the initial one. We illustrate the improvements of our training procedure with experiments using real world datasets for regression tasks under an adversarial setting. We show numerically that solving our proposed convex program yields networks with lower objective values on the Lipschitz-regularized program compared to existing methods. Additionally, we show that on certain datasets, networks obtained using our convex training program are both more accurate and robust with respect to adversarial attacks.
2026-06-16
arXiv (prépublication)
doi.org
arxiv.org
Observability and Consistency Analysis for Visual-Inertial Navigation with Anchored Feature Parameterizations
Mitchell Cohen
Vassili Korotkine
James Richard Forbes
This paper presents an analysis of the observability and consistency properties of filtering-based visual-inertial navigation systems (VINS)… (voir plus) that utilize anchored feature representations. The unobservable subspace of VINS with anchored landmark parameterizations is shown to be independent of the estimated landmark state, which leads to improved estimator consistency properties without any additional modifications. However, the unobservable subspace is still found to depend on the estimated navigation state, necessitating additional consistency-enforcing techniques. Two methods to improve the consistency of VINS with anchored feature representations are presented. Simulation results showcase that all estimators employing anchored feature paramterizations exhibit improved consistency properties compared to algorithms that estimate features resolved in a global reference frame, especially in scenarios where feature initialization may be poor. Real-world experiments on the TUM-VI dataset showcase that the use of anchored feature representations alone can yield comparable performance to consistency-improved estimators employing a global feature representation, demonstrating the benefit of using anchored feature parameterizations for VINS.
2026-06-16
arXiv (prépublication)
doi.org
arxiv.org
Scaling Self-Play for End-to-End Driving
Luke Rowe
Roger Girgis
Rodrigue De Schaetzen
Daphne Cornelisse
Alaap Grandhi
Felix Heide
Eugene Vinitsky
Christopher Pal
Liam Paull
End-to-end autonomous driving models are typically trained on offline human-demonstration datasets that provide limited state coverage and o… (voir plus)ften no closed-loop feedback, making them prone to compounding errors when deployed in closed-loop and brittle to long-tail agent interactions. To overcome these limitations, we propose an alternative strategy for training end-to-end driving models: large-scale self-play directly from pixels in simulation. While prior self-play approaches have shown promising transfer to real-world driving, they typically assume vectorized Bird's-Eye-View (BEV) observations that are incompatible with end-to-end policies operating directly on sensor observations. To this end, we introduce Gigapixel, a high-throughput batched driving simulator with perspective rendering, enabling scalable self-play directly from pixel observations. Rather than targeting compute-costly photorealistic sensor simulation, Gigapixel renders a simplified bounding-box world that preserves essential scene structure while achieving throughput at 50k agent steps per second. Since direct pixel-space self-play RL is prohibitively sample-inefficient at end-to-end model scale, we propose self-play DAgger training: we train pixel-based policies in self-play via on-policy distillation from a privileged RL teacher. To bridge the sim-to-real gap, we subsequently transfer the self-play trained policies to real-world sensor data through lightweight perception adaptation. Policies trained in Gigapixel and adapted to real-world sensor data achieve competitive performance on the HUGSIM and NAVSIM-v2 benchmarks without human trajectory supervision. Moreover, scaling self-play training yields proportional gains in policy performance, establishing self-play as a practical and scalable strategy for training end-to-end models.
2026-06-16
arXiv (prépublication)
doi.org
arxiv.org
Supp Fig1+ Legend from PGV001, a Multi-Peptide Personalized Neoantigen Vaccine Platform: Phase I Study in Patients with Solid and Hematologic Malignancies in the Adjuvant Setting
Mansi Saxena
Thomas U. Marron
Julia Kodysh
John P. Finnigan
Sayali Onkar
Anna Kaminska
Kevin Tuballes
Ruiwei Guo
Rachel Lubong Sabado
Marcia Meseck
Timothy J. O’Donnell
Robert Sebra
Samir Parekh
Matthew D. Galsky
Ana Blasquez
Gustavo Gimenez
Mesude Bicak
Cansu Cimen Bozkus
Daniela Delbeau-Zagelbaum
Denise Rodriguez … (voir 17 de plus)
Ana Acuña-Villaorduña
Krzysztof Misiukiewicz
Marshall R. Posner
Brett A. Miles
Hanna Y. Irie
Amy Tiersten
Deborah B. Doroshow
Andrea Wolf
John Mandeli
Rachel Brody
Andres Μ. Salazar
Sacha Gnjatic
Jeff Hammerbacher
Eric E. Schadt
Philip Friedlander
Alex Rubinsteyn
Nina Bhardwaj
<p>Supplementary Fig1. PGV001 feasibility and recurrence free survival a) Survival plot depicting overall survival (OS). N=12. Median … (voir plus)survival depicted by vertical dotted line b) Survival plot depicting recurrence free survival (RFS). Median RFS is depicted by vertical dotted line. N=12. c) Swimmer plot depicting time-line of clinical events for each patient since their curative intent treatment. d) Comparing linear correlation between TMB or neoantigen load vs OS at 60 months from 1st vaccine. Pearson correlation coefficient (r) calculated for fitting of the correlation with 95% confidence interval (95%CI). TMB: tumor mutation burden, NeoAg: neoantigens, OS: overall survival,</p>
2026-06-16
American Association for Cancer Research (accepté)
doi.org
Supp Fig2 + Legend from PGV001, a Multi-Peptide Personalized Neoantigen Vaccine Platform: Phase I Study in Patients with Solid and Hematologic Malignancies in the Adjuvant Setting
Mansi Saxena
Thomas U. Marron
Julia Kodysh
John P. Finnigan
Sayali Onkar
Anna Kaminska
Kevin Tuballes
Ruiwei Guo
Rachel Lubong Sabado
Marcia Meseck
Timothy J. O’Donnell
Robert Sebra
Samir Parekh
Matthew D. Galsky
Ana Blasquez
Gustavo Gimenez
Mesude Bicak
Cansu Cimen Bozkus
Daniela Delbeau-Zagelbaum
Denise Rodriguez … (voir 17 de plus)
Ana Acuña-Villaorduña
Krzysztof Misiukiewicz
Marshall R. Posner
Brett A. Miles
Hanna Y. Irie
Amy Tiersten
Deborah B. Doroshow
Andrea Wolf
John Mandeli
Rachel Brody
Andres Μ. Salazar
Sacha Gnjatic
Jeff Hammerbacher
Eric E. Schadt
Philip Friedlander
Alex Rubinsteyn
Nina Bhardwaj
<p>Supplementary Fig2. PGV001-induced T cell immunity as measured by ex vivo IFNγ ELISPOT assay PBMC samples were stimulated with neo… (voir plus)antigen peptides (pools of peptides: composed of 15mer OLPs+ 9-10mer predicted Mins corresponding to each vaccine SLP) and analyzed by IFNγ ELISPOT. All data is MOG normalized. a) Line diagrams showing longitudinal changes in IFNγ secretion upon stimulation with responder neoantigen peptides in each patient. Each line represents an individual neoantigen and each dot represents an individual time-point. For definition of “Responder” neoantigen: See Materials and Methods. b) Aligned dot plot showing IFNγ secretion at baseline (Pre) induced by all 126 neoantigens used in the study for 13 patients. Horizontal dotted line depicts cut-off threshold for immunogenicity in the assay. Peptides with immunogenicity at baseline, above the dotted line, are labeled. “Immunogenic” neoantigen: See Materials and Methods. c) Line plot showing post-vaccination immune response, as measured by IFNγ release in ex vivo ELISPOT assay, by the neoantigen peptides found to be “immunogenic” at baseline in (b). d) Plot showing number of IFNγ SFCs/million PBMCs elicited by tetanus peptide at various time points in 13 patients. e) Line plots showing immune response, as measured by IFNγ release in ex vivo ELISPOT assay, elicited by mutated neoantigen peptide pools vs their wild type (WT) counterparts at Week 28 over a range of peptide concentrations. Note: For PID:016 Week 31 sample was used. f) Kaplan Meier Curve comparing OS between subjects that responded, in ex vivo ELISPOT assay, to more or less than 40% neoantigens in their vaccines. PID-017 was lost of follow up and excluded from this analysis. g) Comparing number of SFCs/million PBMCs elicited by each neoantigen in patients Alive (N=6) or Deceased (N=4) at 60-month survival follow up. Total 96 neoantigens analyzed. h) Pie chart showing proportion of neoantigens that elicited antigen specific response in patients within Alive (N=6) vs Deceased (N=4) cohorts. SFC: spot forming cells. In f-h patients that expired with no evidence of their disease recurrence are excluded. *p value indicates two-sided Student’s T-test. *** <0.001. # p value indicates Log-rank (Matel-Cox) test ### <0.001. in patients within Alive (N=6) vs Deceased (N=4) cohorts. SFC: spot forming cells. In f-h patients that expired with no evidence of their disease recurrence are excluded. *p value indicates two-sided Student’s T-test. *** <0.001. # p value indicates Log-rank (Matel-Cox) test ### <0.001.</p>
2026-06-16
American Association for Cancer Research (accepté)
doi.org
Supp Fig3 + Legend from PGV001, a Multi-Peptide Personalized Neoantigen Vaccine Platform: Phase I Study in Patients with Solid and Hematologic Malignancies in the Adjuvant Setting
Mansi Saxena
Thomas U. Marron
Julia Kodysh
John P. Finnigan
Sayali Onkar
Anna Kaminska
Kevin Tuballes
Ruiwei Guo
Rachel Lubong Sabado
Marcia Meseck
Timothy J. O’Donnell
Robert Sebra
Samir Parekh
Matthew D. Galsky
Ana Blasquez
Gustavo Gimenez
Mesude Bicak
Cansu Cimen Bozkus
Daniela Delbeau-Zagelbaum
Denise Rodriguez … (voir 17 de plus)
Ana Acuña-Villaorduña
Krzysztof Misiukiewicz
Marshall R. Posner
Brett A. Miles
Hanna Y. Irie
Amy Tiersten
Deborah B. Doroshow
Andrea Wolf
John Mandeli
Rachel Brody
Andres Μ. Salazar
Sacha Gnjatic
Jeff Hammerbacher
Eric E. Schadt
Philip Friedlander
Alex Rubinsteyn
Nina Bhardwaj
<p>Supplementary Fig3. Neoantigen specific antibody responses induced by PGV001. Patient plasma was subjected to seromics by ELISA usi… (voir plus)ng linear SLPs in the vaccines. N=12 patients. PID-017 excluded from analysis due to high background. a) Stacked bar graph depicting number of neoantigens that induced an IgG/A or M response in each patient at Week 8 and Week 28. For seromics by ELISA a “Responder” neoantigen is defined as a peptide that induced an antibody titer of greater than 100 compared to baseline. b) Line diagrams showing changes in antibody isotypes induced by responder peptides in patients from pre-treatment (Pre), through Prime (Week 8) and Post (Week 28, Week 31 or End of treatment (EOT)). c) Line graph showing changes in titers of total IgG-subclasses induced by responder peptides in PID-006 and PID-008. d) Heat map depicting poly-ICLC specific antibody responses in each patient. e) Plot depicting linear correlation between total IgG antibody titer and IFNγ ELISPOT response at Week 8 (Prime) and Week 28/31 (Post). Pearson correlation coefficient and Spearman correlation coefficient calculated for fitting of the correlation with 95% confidence interval (95%CI). Each dot represents a neoantigen. Horizontal line depicts threshold for antibody titer response at 100. The vertical dotted line depicts threshold for ELISPOT immunogenicity at 60 SFC/million PBMCs. For Week 8, 126 neoantigens evaluated while for Week 28/31, 116 neoantigens were evaluated.</p>
2026-06-16
American Association for Cancer Research (accepté)
doi.org
Supp Fig4+ Legend from PGV001, a Multi-Peptide Personalized Neoantigen Vaccine Platform: Phase I Study in Patients with Solid and Hematologic Malignancies in the Adjuvant Setting
Mansi Saxena
Thomas U. Marron
Julia Kodysh
John P. Finnigan
Sayali Onkar
Anna Kaminska
Kevin Tuballes
Ruiwei Guo
Rachel Lubong Sabado
Marcia Meseck
Timothy J. O’Donnell
Robert Sebra
Samir Parekh
Matthew D. Galsky
Ana Blasquez
Gustavo Gimenez
Mesude Bicak
Cansu Cimen Bozkus
Daniela Delbeau-Zagelbaum
Denise Rodriguez … (voir 17 de plus)
Ana Acuña-Villaorduña
Krzysztof Misiukiewicz
Marshall R. Posner
Brett A. Miles
Hanna Y. Irie
Amy Tiersten
Deborah B. Doroshow
Andrea Wolf
John Mandeli
Rachel Brody
Andres Μ. Salazar
Sacha Gnjatic
Jeff Hammerbacher
Eric E. Schadt
Philip Friedlander
Alex Rubinsteyn
Nina Bhardwaj
<p>Supplementary Fig4. CD4+ and CD8+ T cell activation by PGV001 PBMCs from patients were expanded in vitro in presence of neoantigen … (voir plus)peptides (pools of peptides, OLPs+Mins, corresponding to each vaccine SLP) and restimulated with peptide/s followed by intracellular staining and flowcytometry. All data is normalized to MOG. a) Aligned dot plot showing CD8+ and CD4+ T cell responses at baseline induced by 96 evaluated neoantigens across 10 patients. Horizontal dotted line depicts cut-off threshold for immunogenicity. 15 and 32 peptides were found to be immunogenic at baseline against CD8+ and CD4+ T cells, respectively. b) Line plots showing longitudinal CD8+ and CD4+ T cell immune responses induced by neoantigens found to be immunogenic at baseline in (a) and that induced greater than 2-fold increase in response following PGV001 treatment. c) Each pie chart shows proportion of vaccine peptides that induced production of IFNγ, TNFα or IL-2 in CD8+ or CD4+T cells in each patient at any timepoint post vaccine initiation. Refer to Material and Methods for definition of “response” in flowcytometry assay. d) CD8+ and CD4+ T cell immune responses elicited by tetanus peptides at various time points in 10 evaluated patients.</p>
2026-06-16
American Association for Cancer Research (accepté)
doi.org
Supp Fig5+ Legend from PGV001, a Multi-Peptide Personalized Neoantigen Vaccine Platform: Phase I Study in Patients with Solid and Hematologic Malignancies in the Adjuvant Setting
Mansi Saxena
Thomas U. Marron
Julia Kodysh
John P. Finnigan
Sayali Onkar
Anna Kaminska
Kevin Tuballes
Ruiwei Guo
Rachel Lubong Sabado
Marcia Meseck
Timothy J. O’Donnell
Robert Sebra
Samir Parekh
Matthew D. Galsky
Ana Blasquez
Gustavo Gimenez
Mesude Bicak
Cansu Cimen Bozkus
Daniela Delbeau-Zagelbaum
Denise Rodriguez … (voir 17 de plus)
Ana Acuña-Villaorduña
Krzysztof Misiukiewicz
Marshall R. Posner
Brett A. Miles
Hanna Y. Irie
Amy Tiersten
Deborah B. Doroshow
Andrea Wolf
John Mandeli
Rachel Brody
Andres Μ. Salazar
Sacha Gnjatic
Jeff Hammerbacher
Eric E. Schadt
Philip Friedlander
Alex Rubinsteyn
Nina Bhardwaj
<p>Supplementary Fig5. TCR sequencing on neoantigen reactive T cells in patient PBMCs a) (Top) Depiction of methodology utilized to ge… (voir plus)nerate ex vivo and in vitro expanded and stimulated (IVS) T cell samples from screen, week 8 (W8) and week 28 (W28) timepoints for TCR sequencing. (Bottom) Gating strategy and representative dot plot for isolation by FAC sorting of IFNγ positive, neoantigen reactive T cells for TCR sequencing. b) (Top) Representative differential abundance plots from PID:012 comparing TCR clones between two conditions listed on X and Y axes. (Bottom) Logic and calculation of “relevant” clones in vivo. First using the IVS T cells, lists of TCR sequences differentially abundant in peptide stimulated (Pepstim) Week 8 (W8) and Week 28 (W28) versus screening were generated, named W8IVS>scr and W28IVS>scr, respectively. Similarly, lists of TCR sequences differentially abundant in peptide expanded W8 and W28 samples versus DMSO expanded cells was obtained, named W8ctrl>IVS and W28ctrl>IVS, respectively. Sequences in W8ctrl>IVS and W28ctrl>IVS were removed from W8IVS>scr and W28IVS>scr to obtain vaccine induced “relevant” TCR sequences in IVS T cells named, W8ΔIVS and W28ΔIVS. Ex vivo T cells were analyzed to identify lists of differentially abundant TCRs in W8 and W28 versus screening, named W8ex>scr and W28ex>scr. Finally, an intersection between relevant clones in IVS (W8ΔIVS and W28ΔIVS) and abundant clones in W8 or W28 ex vivo T cells (W8ex>scr and W28ex>scr) was performed to identify relevant neoantigen reactive TCR sequences in vivo, named W8clones and W28clones. c) Stacked bar plot depicting number and frequency of new (detected only post vaccination) and expanded (existed at baseline but frequency expanded post vaccination) W8clones and W28clones TCR clones in PID:015. Each color represents distinct TCR clones and the height of each stack represents each clone’s frequency. d) For PID:015, Venn diagram showing overlap between ex vivo and IVS clones to obtain W8clones and W28clones and their overlap to identify persisting clones in vivo. For PID:015, no persisting clones were found.</p>
2026-06-16
American Association for Cancer Research (accepté)
doi.org