Publications

One Image Is Enough: Self-Generated Supervision for Personalizing Human Video Diffusion at Test Time
Sadman Rakib Pinon
Yanis Falaki
Xinxin Zuo
Pose-driven human video diffusion animates a person from a single reference image and a target pose sequence, yet it struggles to preserve i… (see more)dentity, rendering the face and garments unfaithfully as the pose departs from it. Personalization is the standard remedy, but its supervision is missing: one image fixes identity at a single pose, while a multi-pose subject video cannot be assumed in practice. Instead, we synthesize the missing supervision. A domain-tuned, pose-conditioned generator produces pseudo-ground-truth frames at selected key poses. An inference-time reference-only sampler steers these candidates toward the subject and keeps the most identity-consistent. A per-subject low-rank adapter then trains on static clips of them, configured from the reference alone, with no subject video, ground truth, or labels. Although lower in fidelity, these synthetic frames supervise more effectively than real ones. Real frames as captured carry motion that erodes the temporal prior, while a static clip conveys appearance alone. On TikTok and UBC Fashion, the adapted model improves the base on every metric, inverts the identity-versus-motion trade-off, and matches or surpasses adaptation on five real frames despite using none. Code and weights are available at https://github.com/SadmanPinon/one-image-is-enough.
OpenZilo: An Open Smart-Ring Platform for Human-Centric Sensing and Interaction with Robots
Xingyao Wang
Rui Feng
Chi Qin
Weijian Lin
Lizhi Ding
Jiankai Tang
Yuntao Wang
Xiangke Liu
Wearable robot interfaces must connect user input to timely physical action. We present the proposed platform, an open smart-ring platform f… (see more)or studying this connection across task stages. The ring integrates a microphone, an inertial measurement unit, photoplethysmography, and skin-temperature sensing. Its software architecture separates acquisition, event interpretation, interaction policy, and robot execution. We characterize seven operating modes and evaluate speech and gesture control with 10 participants. Across five commands and 250 trials per modality, speech and gestures achieved task success rates of 90.4% and 88.8%, respectively. Recognition accuracy exceeded task success for both modalities. Gaussian hidden Markov models achieved higher accuracy than a support vector machine across three vocabulary sizes, with longer scoring times. Following a shared spoken task start, gesture-based adjustments had shorter median response times than speech. Whole-task success was 70.0% with gestures and 64.0% with speech. We further examine physiological-event-triggered robot interaction using simulated event notifications. Mean notification-receipt-to-arrival times were 11.71 s for the robot and 9.20 s for an on-site person. These descriptive findings support evaluating wearable robot interfaces through both recognition and physical outcomes, with timing measures appropriate to each task stage.
OpenZilo: An Open Smart-Ring Platform for Human-Centric Sensing and Interaction with Robots
Xingyao Wang
Rui Feng
Chi Qin
Weijian Lin
Lizhi Ding
Jiankai Tang
Yuntao Wang
Xiangke Liu
Wearable robot interfaces must connect user input to timely physical action. We present the proposed platform, an open smart-ring platform f… (see more)or studying this connection across task stages. The ring integrates a microphone, an inertial measurement unit, photoplethysmography, and skin-temperature sensing. Its software architecture separates acquisition, event interpretation, interaction policy, and robot execution. We characterize seven operating modes and evaluate speech and gesture control with 10 participants. Across five commands and 250 trials per modality, speech and gestures achieved task success rates of 90.4% and 88.8%, respectively. Recognition accuracy exceeded task success for both modalities. Gaussian hidden Markov models achieved higher accuracy than a support vector machine across three vocabulary sizes, with longer scoring times. Following a shared spoken task start, gesture-based adjustments had shorter median response times than speech. Whole-task success was 70.0% with gestures and 64.0% with speech. We further examine physiological-event-triggered robot interaction using simulated event notifications. Mean notification-receipt-to-arrival times were 11.71 s for the robot and 9.20 s for an on-site person. These descriptive findings support evaluating wearable robot interfaces through both recognition and physical outcomes, with timing measures appropriate to each task stage.
TextBand: Writing by Visualizing and Editing the Sound of Text
Damien Masson
Traditional writing tools focus on how text looks, not how it sounds. We introduce TextBand, a writing tool that helps edit the sonority and… (see more) musicality of text using a design inspired by music composition software: text is on a timeline to help visualize and edit rhythm; sentences can be heard through text-to-speech playback; and sonorities can be viewed and edited on a phonetic grid similar to a MIDI editor. Results from a user study showed that TextBand helps identify 35% more sonority patterns and successfully fix 57% more sonority issues while helping consciously use figures of speech such as assonance and alliteration. Our work challenges the design of existing writing software and paves the way for more sound-based support when writing.
The Association of MRI-Derived Tissue Bridge Morphometrics with Baseline Impairment and Recovery in Traumatic Spinal Cord Injury: A Subanalysis of the RISCIS Randomized Controlled Trial
Mouad Elganga
Mohammed Ali Alvi
Karlo Miguel Pedro
Yuxi Zhang
Julien Cohen‐Adad
Michael Hardisty
Shekar N. Kurpad
Aditya Vedantam
Bizhan Aarabi
Wilson Zack Ray
Michael George Fehlings
Early prognostic biomarkers are essential to optimize treatment and trial design in spinal cord injury (SCI). The Riluzole in SCI Study (RIS… (see more)CIS), a multicenter randomized controlled trial, provided evidence on secondary analyses that riluzole improved functional recovery in acute cervical SCI. The current analysis uses magnetic resonance imaging (MRI) to assess lesion and tissue bridge morphometrics against baseline impairment, functional recovery, and treatment response in RISCIS participants. In this preplanned ambispective subanalysis, 24 participants underwent T2-weighted MRI for acute cervical SCI together with neurological and functional assessments to 12 months. A subset also had diffusion tensor imaging (DTI), used to extract fractional anisotropy (FA) values. Segmentation with the Spinal Cord Toolbox (v7.2) gave ventral, dorsal, and total tissue-bridge widths and lesion volume and length. Associations with American Spinal Injury Association Impairment Scale (AIS) grade, upper extremity motor score (UEMS) and total motor score (TMS), lower extremity motor score (LEMS), light touch and pinprick sensory scores, Spinal Cord Independence Measure (SCIM), and Short Form-36 (SF-36) domains at baseline and follow-up were analyzed with regression models adjusted for age and sex. Overall recovery across TMS, SCIM, and SF-36 Physical Component Summary was analyzed with an O’Brien-style global statistical test (GST). Interactions tested whether bridge metrics modified riluzole’s effect. At baseline, greater ventral ( β = 9.65, p = 0.017) and total ( β = 6.16, p = 0.021) tissue-bridge widths were associated with higher TMS, and ventral width with higher UEMS ( β = 6.61, p = 0.010). Larger dorsal (odds ratio [OR] = 21.92, p = 0.013), ventral (OR = 4.91, p = 0.049), and total (OR = 3.70, p = 0.019) bridges corresponded to less severe baseline AIS grade. Pinprick scores increased with ventral ( β = 26.56, p = 0.020) and total ( β = 17.50, p = 0.023) bridge width, and light touch scores increased with total bridge width ( β = 16.22, p = 0.045). No imaging metric was associated with LEMS. These individual-metric associations were nominally significant but did not survive correction. In GST analyses, larger ventral ( T = 4.63, p = 0.012, p ′= 0.048) and total ( T = 4.30, p = 0.019, p ′=0.048) bridges predicted greater overall recovery; both remained significant after correction and were stable across sensitivity analyses. Ventral and total bridge width showed nominally significant interactions with riluzole treatment for upper-limb motor recovery. Exploratory DTI findings were inconclusive and sensitive to individual observations: both the FA–recovery association and the observation of poor outcome despite a preserved tissue bridge rested on a single participant. Overall, we found that midsagittal tissue-bridge width measured ∼72 h after injury from standard-of-care T2-weighted MRI was associated with baseline injury severity and multidomain functional recovery, at a timepoint substantially earlier than the 1 month at which tissue bridges have previously been quantified. Because edema may obscure lesion margins this early, acute bridge widths may underestimate spared tissue resolvable later. The treatment-interaction observation is hypothesis-generating and requires prospective confirmation. These findings support MRI-derived tissue bridges as candidate biomarkers for prognostication and enrichment of neuroprotective trials in SCI.
Adaptive Preference Elicitation in Preference Robust CPT-Based Shortfall
Sainan Zhang
Huifu Xu
This paper investigates the cumulative prospect theory (CPT)-based shortfall risk measure where the value (utility) function and the distort… (see more)ion (probability weighting) functions are ambiguous. We propose a worst-case CPT-based shortfall risk measure where the capital requirement is based on the worst-case value function and distortion functions from an ambiguity set constructed with user-specific information (e.g., preference between risky and certain lotteries) and well-known general information such as the S-shape of the value functions. When the underlying random loss is discretely distributed and parameters in pairs of lotteries remain undetermined, we demonstrate that the worst-case value function and the distortion functions have a piecewise linear structure. Based on this observation, we develop an adaptive preference elicitation approach extended from the well-known polyhedral method to simultaneously elicit the distortion functions and the positive and negative parts of the value function by solving a sequence of mixed-integer linear programs. Based on the elicited distortions, we design new queries to construct a feasible set of the holistic value functions. The proposed framework iteratively shrinks the ambiguity set through designed queries, yielding a computationally implementable robust CPT-based shortfall risk measure. To examine the in-sample and out-of-sample performance of the elicited CPT-based preference functions and the resulting preference robust shortfall risk measure, we conduct numerical experiments under an S-shaped value function and inverse S-shaped distortion functions and report the corresponding results.
Beyond the ICP value: neurocritical care in severe bacterial meningitis- Authors’ reply
Cindy Poirier
Benoit Champigneulle
Cross-component causal coupling modeling and multi-order spectral representation for interpretable mechanical fault diagnosis
Yi Gao
Haidong Shao
Jiarui Liu
Mingze Xu
Jiale Zhang
Daolin Xu
DailyWildfireCAN: A Dataset and Benchmark for Next-Day Wildfire Spread Prediction in Canada
Mahacine Ettahri
Hager Radi Abdelwahed
Daniel Aloise
Wildfires represent a critical ecological and socio-economic challenge in Canada, where vast boreal forests and a warming climate have ampli… (see more)fied the frequency and severity of extreme fire seasons. Predicting next-day wildfire spread can support fire management, but requires capturing complex interactions among weather, fuels, and terrain. Deep learning has emerged as a promising approach to learning these relationships from heterogeneous environmental data, yet its development and evaluation depend on representative datasets. Existing deep learning research on wildfire spread relies predominantly on US-focused datasets, leaving Canadian fire environments underrepresented. To address this gap, we present DailyWildfireCAN, a multi-modal, multi-temporal dataset and benchmark for predicting next-day wildfire spread across Canada. The dataset tracks continuous fire progression over multi-day events at a high spatial and temporal resolution of 90-m and 24-hour, respectively. This resolution preserves fine-scale information relevant to fire-edge geometry, local fuel variations, and terrain effects, supporting research toward improved daily wildfire spread forecasting. DailyWildfireCAN covers 2,440 wildfire events across eight fire seasons 2017-2024 and integrates 33 features across diverse modalities, including topography, meteorology, land cover, vegetation and forest structure, human influence indices, and satellite imagery. We establish a comprehensive benchmark comparing mono-temporal and multi-temporal prediction models, accompanied by a multi-dimensional evaluation framework assessing robustness across wildfire characteristics, geographic regions, and temporal conditions, spatial prediction quality, and sensitivity to reference label variations.
HEAMS: Holistic Evaluation of Agent Memory Systems
Memory systems allow AI agents to persist information across interactions, making them more personalized and more capable on long-horizon ta… (see more)sks. This persistence introduces new risks: stored information can be disclosed in inappropriate contexts, resist correction or deletion, be retained beyond what users desire, or be poisoned to influence later agent behaviour. Yet memory systems are typically evaluated on task performance and efficiency alone. We argue that a memory system should instead be understood as the custodian of its user's persistent information, with responsibilities that extend beyond helping an agent complete its tasks, and that it should be evaluated holistically against all such responsibilities. We present Holistic Evaluation of Agent Memory Systems (HEAMS), an open-source framework for evaluating agent memory systems in this way. First, we state the responsibilities of agent memory systems as explicit behavioural desiderata spanning performance, efficiency, security, privacy, and user sovereignty. Second, we assemble a benchmark suite that covers these desiderata and acknowledge the gaps that remain. Third, we develop MilaMem, a prototype memory system, and a governance layer that can wrap any memory system, enabling controlled study of memory system design choices. Fourth, we compare memory systems across the full HEAMS benchmark suite. We find that no memory system leads every performance benchmark, and strong performance scores may coexist with poor privacy, user-sovereignty, and security scores. Governance strongly influences results: recall policies can reduce inappropriate disclosure at a cost in helpfulness, while both recall and retention policies improve poisoning resistance. Performance and efficiency benchmarks alone therefore cannot reveal how well a memory system meets its responsibilities. Code: https://github.com/milatechtransfer/heams
A reproducible pipeline for simulating individual and inter-brain neural dynamics from structural connectivity
Annemarie Wolff
Mahta Ramezanian Panahi
Imme Christina Schultheiß
Juha M. Lahnakoski
Kai Vogeley
Marc Tittgemeyer
Leonhard Schilbach
Tactile Sensing via Sparse Pressure Measurements in a Suspension
Wilfred Mason
Omar Khater
Charles Sirois
Tactile sensors with few transducers can resolve rich dynamics in soft robots, haptic interfaces, and wearable devices with minimal added co… (see more)mplexity. Using only four transducers, we design a sensor to infer position, depth, and duration of a touch input through the dissipative properties of a granular suspension. Though many tactile sensors rely on learned features to infer input characteristics, we show that models trained on simple time and pressure features perform comparatively to a convolutional neural network baseline. We extract pressure amplitude, arrival time delay, and pulse width from pressure sensor time series to estimate touch characteristics. Then, we fit interpolation functions for press position and press time as a function of pressure amplitude, relative peak time, and press width. With this approach, we directly obtain the press location and time from time series measurements. On average, the proposed sensor estimates press position within 10% with a standard deviation of 7% and press time within 5.9% with a standard deviation of 11%. Additionally, this sensor provides reasonable prediction performance with as few as two pressure transducers at various concentrations, probe sizes, and for different boundary conditions. To showcase the simplicity of this sensing and estimation scheme, we demonstrate its implementation on different geometries and embedded hardware in real-time. The rich dynamics of this sensor may further enable gesture and shape detection, ultimately enabling closed-loop feedback for granular haptic interfaces and soft robots.