Subpopulation shift, characterized by a disparity in subpopulation distribution between the training and target datasets, can significantly degrade the performance of machine learning models. Current solutions to subpopulation shift involve modifying empirical risk minimization with re-weighting strategies to improve generalization. This strategy relies on assumptions about the number and nature of subpopulations and annotations on group membership, which are unavailable for many real-world datasets. Instead, we propose using an ensemble of diverse classifiers to adaptively capture risk associated with subpopulations. Given a feature extractor network, we replace its standard linear classification layer with a mixture of prototypical classifiers, where each member is trained to classify the data while focusing on different features and samples from other members. In empirical evaluation on nine real-world datasets, covering diverse domains and kinds of subpopulation shift, our method of Diverse Prototypical Ensembles (DPEs) often outperforms the prior state-of-the-art in worst-group accuracy. The code is available at https://github.com/minhto2802/dpe4subpop.

This paper presents a machine-learning (ML)-assisted co-design framework for the optimization of open-loop analog linearization in VCO-based ADCs. A deep neural network (DNN) surrogate model is trained on a dataset generated by transistor-level transient simulations of the VCO’s voltage-to-frequency (V -to-f) characteristic. The vast multi-dimensional optimization landscape provided by the VCO’s embedded linearization tuning ‘knobs’ make exhaustive search infeasible. The DNN model enables a rapid exploration of this tuning-knobs landscape to minimize the harmonic-distortion (HD) through an evolutionary genetic algorithm (GA). The ‘predicted’ optimal tuning-knob values are transferred back into the local Cadence simulation environment for further fine-grained gradient-descent adaptation via an integrated Verilog-A self-calibration engine. Applied to the coupled-oscillator-ensemble (COE) circuit architecture, the tuning-knob configurations inferred by the DNN-GA yield simulated third-order HD of 42–52dB for the VCO-based ADC, with a boost to >60dB (i.e., 10-bit performance) using a few iterations of local gradient-descent optimization.

Shortcut models represent a promising, non-adversarial paradigm for generative modeling, uniquely supporting one-step, few-step, and multi-step sampling from a single trained network. However, their widespread adoption has been stymied by critical performance bottlenecks. This paper tackles the five core issues that held shortcut models back: (1) the hidden flaw of compounding guidance, which we are the first to formalize, causing severe image artifacts; (2) inflexible fixed guidance that restricts inference-time control; (3) a pervasive frequency bias driven by a reliance on low-level distances in the direct domain, which biases reconstructions toward low frequencies; (4) divergent self-consistency arising from a conflict with EMA training; and (5) curvy flow trajectories that impede convergence. To address these challenges, we introduce iSM, a unified training framework that systematically resolves each limitation. Our framework is built on four key improvements: Intrinsic Guidance provides explicit, dynamic control over guidance strength, resolving both compounding guidance and inflexibility. A Multi-Level Wavelet Loss mitigates frequency bias to restore high-frequency details. Scaling Optimal Transport (sOT) reduces training variance and learns straighter, more stable generative paths. Finally, a Twin EMA strategy reconciles training stability with self-consistency. Extensive experiments on ImageNet 256x256 demonstrate that our approach yields substantial FID improvements over baseline shortcut models across one-step, few-step, and multi-step generation, making shortcut models a viable and competitive class of generative models.

The distribution of data changes over time; models operating in dynamic environments need retraining. But knowing when to retrain, without access to labels, is an open challenge since some, but not all shifts degrade model performance. This paper formalizes and addresses the problem of post-deployment deterioration (PDD) monitoring. We propose D3M, a practical and efficient monitoring algorithm based on the disagreement of predictive models, achieving low false positive rates under non-deteriorating shifts and provides sample complexity bounds for high true positive rates under deteriorating shifts. Empirical results on both standard benchmark and a real-world large-scale internal medicine dataset demonstrate the effectiveness of the framework and highlight its viability as an alert mechanism for high-stakes machine learning pipelines.

Denoising Probabilistic Models (DPMs) represent an emerging domain of generative models that excel in generating diverse and high-quality images. However, most current training methods for DPMs often neglect the correlation between timesteps, limiting the model's performance in generating images effectively. Notably, we theoretically point out that this issue can be caused by the cumulative estimation gap between the predicted and the actual trajectory. To minimize that gap, we propose a novel sequence-aware loss that aims to reduce the estimation gap to enhance the sampling quality. Furthermore, we theoretically show that our proposed loss function is a tighter upper bound of the estimation loss in comparison with the conventional loss in DPMs. Experimental results on several benchmark datasets including CIFAR10, CelebA, and CelebA-HQ consistently show a remarkable improvement of our proposed method regarding the image generalization quality measured by FID and Inception Score compared to several DPM baselines. Our code and pre-trained checkpoints are available at https://github.com/VinAIResearch/SA-DPM.

We study the problem of online sequential decision-making given auxiliary demonstrations from _experts_ who made their decisions based on unobserved contextual information. These demonstrations can be viewed as solving related but slightly different tasks than what the learner faces. This setting arises in many application domains, such as self-driving cars, healthcare, and finance, where expert demonstrations are made using contextual information, which is not recorded in the data available to the learning agent. We model the problem as a zero-shot meta-reinforcement learning setting with an unknown task distribution and a Bayesian regret minimization objective, where the unobserved tasks are encoded as parameters with an unknown prior. We propose the Experts-as-Priors algorithm (ExPerior), an empirical Bayes approach that utilizes expert data to establish an informative prior distribution over the learner's decision-making problem. This prior enables the application of any Bayesian approach for online decision-making, such as posterior sampling. We demonstrate that our strategy surpasses existing behaviour cloning and online algorithms, as well as online-offline baselines for multi-armed bandits, Markov decision processes (MDPs), and partially observable MDPs, showcasing the broad reach and utility of ExPerior in using expert demonstrations across different decision-making setups.

A frequency-to-digital converter (FDC) performs the role of precise frequency digitization within a voltage-controlled oscillator (VCO)-based ADC. To be compatible with energy-harvesting (EH) Internet-of-Things (IoT) devices, the development of ultra-low-voltage (ULV) FDCs is crucial, where the primary focus must be directed towards the maximization of data throughput under dramatic constraints of reliability and timing variability associated with deep-subthreshold operation. This article investigates the speed maximization of a 0.2V full-custom ULV FDC design, consisting of an array of several parallel XOR-based FDC units, and the multi-rate decimation-filtering digital back-end. At the core of this broad exploration is a high-speed sense-amplify phase sampler (PS) featuring hardware redundancy, capable of sampling the phase of low-voltage-swing inputs. Particular focus is placed on the yield-based reliability-driven design methodology for the sense-amplify phase-sampling circuits running up to 40MS/s and practical variability-mitigation strategies. To overcome the speed bottleneck in the digital back-end, a fully parallel bitstream-processing architectural composition of the computations for summation and decimation are proposed. Experimental verification through measurements of the FDC integrated within a 10-bit 160kHz bandwidth (BW) open-loop VCO-based ADC across clock frequency with supply variations demonstrate robust operation of the first 0.2V multi-phase FDC in the advanced 28nm CMOS process.

OBJECTIVES: The HL7® fast healthcare interoperability resources (FHIR®) specification has emerged as the leading interoperability standard for the exchange of healthcare data. We conducted a scoping review to identify trends and gaps in the use of FHIR for clinical research. MATERIALS AND METHODS: We reviewed published literature, federally funded project databases, application websites, and other sources to discover FHIR-based papers, projects, and tools (collectively, "FHIR projects") available to support clinical research activities. RESULTS: Our search identified 203 different FHIR projects applicable to clinical research. Most were associated with preparations to conduct research, such as data mapping to and from FHIR formats (n = 66, 32.5%) and managing ontologies with FHIR (n = 30, 14.8%), or post-study data activities, such as sharing data using repositories or registries (n = 24, 11.8%), general research data sharing (n = 23, 11.3%), and management of genomic data (n = 21, 10.3%). With the exception of phenotyping (n = 19, 9.4%), fewer FHIR-based projects focused on needs within the clinical research process itself. DISCUSSION: Funding and usage of FHIR-enabled solutions for research are expanding, but most projects appear focused on establishing data pipelines and linking clinical systems such as electronic health records, patient-facing data systems, and registries, possibly due to the relative newness of FHIR and the incentives for FHIR integration in health information systems. Fewer FHIR projects were associated with research-only activities. CONCLUSION: The FHIR standard is becoming an essential component of the clinical research enterprise. To develop FHIR's full potential for clinical research, funding and operational stakeholders should address gaps in FHIR-based research tools and methods.

We propose a model-free reinforcement learning algorithm inspired by the popular randomized least squares value iteration (RLSVI) algorithm as well as the optimism principle. Unlike existing upper-confidence-bound (UCB) based approaches, which are often computationally intractable, our algorithm drives exploration by simply perturbing the training data with judiciously chosen i.i.d. scalar noises. To attain optimistic value function estimation without resorting to a UCB-style bonus, we introduce an optimistic reward sampling procedure. When the value functions can be represented by a function class $\mathcal{F}$, our algorithm achieves a worst-case regret bound of $\tilde{O}(\mathrm{poly}(d_EH)\sqrt{T})$ where $T$ is the time elapsed, $H$ is the planning horizon and $d_E$ is the \emph{eluder dimension} of $\mathcal{F}$. In the linear setting, our algorithm reduces to LSVI-PHE, a variant of RLSVI, that enjoys an $\tilde{\mathcal{O}}(\sqrt{d^3H^3T})$ regret. We complement the theory with an empirical evaluation across known difficult exploration tasks.

Driver assistance and vehicular automation would greatly benefit from uninterrupted lane-level vehicle positioning, especially in challenging environments like metropolitan cities. In this paper, we explore whether the WiFi Fine Time Measurement (FTM) protocol, with its robust, accurate ranging capability, can complement current GPS and odometry systems to achieve lane-level positioning in urban canyons. We introduce Wi-Go, a system that simultaneously tracks vehicles and maps WiFi access point positions by coherently fusing WiFi FTMs, GPS, and vehicle odometry information together. Wi-Go also adaptively controls the FTM messaging rate from clients to prevent high bandwidth usage and congestion, while maximizing the tracking accuracy. Wi-Go achieves lane-level vehicle positioning (1.3 m median and 2.9 m 90-percentile error), an order of magnitude improvement over vehicle built-in GPS, through vehicle experiments in the urban canyons of Manhattan, New York City, as well as in suburban areas (0.8 m median and 3.2 m 90-percentile error).

Head-mounted devices (HMD) for Augmented Reality (AR) are gaining traction thanks to a growing number of applications in the areas of image guided therapy, computer aided design, cargo packing, manufacturing and digital field service. However, providing an always available, intuitive and user friendly input for these devices remains a challenging problem. This paper explores recognizing dynamic, micro finger gestures using capacitive coupling for interacting with a head-mounted device. Electrodes are attached to fingertips of users gloves and capacitive coupling among all pairs of electrodes is measured quickly to infer the real-time spatial relationship between fingers. The system is able to recognize fine, low-effort finger gestures, such as swiping, sliding, tap, double-tap. We evaluated our prototype with 14 gestures executed by 10 subjects and found a 97% accuracy of gesture recognition.

This paper explores the feasibility of localizing and detecting activities of building occupants using visible light sensing across a mesh of light bulbs. Existing Visible Light activity sensing (VLS) techniques require either light sensors to be deployed on the floor or a person to carry a device. Our approach integrates photosensors with light bulbs and exploits the light reflected off the floor to achieve an entirely device-free and light source based system. This forms a mesh of virtual light barriers across networked lights to track shadows cast by occupants. The design employs a synchronization circuit that implements a time division signaling scheme to differentiate between light sources and a sensitive sensing circuit to detect small changes in weak reflections. Sensor readings are fed into indoor supervised tracking algorithms as well as occupancy and activity recognition classifiers. Our prototype uses modified off-the-shelf LED flood light bulbs and is installed in a typical office conference room. We evaluate the performance of our system in terms of localization, occupancy estimation and activity classification, and find a 0.89m median localization error as well as 93.7% and 93.78% occupancy and activity classification accuracy, respectively.

Academic and industry research has argued for supporting WiFi time-of-flight measurements to improve WiFi localization. The IEEE 802.11-2016 now includes a Fine Time Measurement (FTM) protocol for WiFi ranging, and several WiFi chipsets offer hardware support albeit without fully functional open software. This paper introduces an open platform for experimenting with fine time measurements and a general, repeatable, and accurate measurement framework for evaluating time-based ranging systems. We analyze the key factors and parameters that affect the ranging performance and revisit standard error correction techniques for WiFi time-based ranging system. The results confirm that meter-level ranging accuracy is possible as promised, but the measurements also show that this can only be consistently achieved in low-multipath environments such as open outdoor spaces or with denser access point deployments to enable ranging at or above 80 MHz bandwidth.

The growing number of devices we interact with require a convenient yet secure solution for user identification, authorization and authentication. Current approaches are cumbersome, susceptible to eavesdropping and relay attacks, or energy inefficient. In this paper, we propose a body-guided communication mechanism to secure every touch when users interact with a variety of devices and objects. The method is implemented in a hardware token worn on user's body, for example in the form of a wristband, which interacts with a receiver embedded inside the touched device through a body-guided channel established when the user touches the device. Experiments show low-power (uJ/bit) operation while achieving superior resilience to attacks, with the received signal at the intended receiver through the body channel being at least 20dB higher than that of an adversary in cm range.

Steerable surveillance cameras offer a unique opportunity to support multiple vision applications simultaneously. However, state-of-art camera systems do not support this as they are often limited to one application per camera. We believe that we should break the one-to-one binding between the steerable camera and the application. By doing this we can quickly move the camera to a new view needed to support a different vision application. When done well, the scheduling algorithm can support a larger number of applications over an existing network of surveillance cameras. With this in mind we developed Panoptes, a technique that virtualizes a camera view and presents a different fixed view to different applications. A scheduler uses camera controls to move the camera appropriately providing the expected view for each application in a timely manner, minimizing the impact on application performance. Experiments with a live camera setup demonstrate that Panoptes can support multiple applications, capturing up to 80% more events of interest in a wide scene, compared to a fixed view camera.

Embedded screen-camera communication techniques encode information in screen imagery that can be decoded with a camera receiver yet remains unobtrusive to the human observer. These techniques have applications in tagging content on screens similar to QR-code tagging for other objects. This paper characterizes the design space for flicker-free embedded screen-camera communication. In particular, we identify an orthogonal dimension to prior work: spatial content-adaptive encoding, and observe that it is essential to combine multiple dimensions to achieve both high capacity and minimal flicker. From these insights, we develop content-adaptive encoding techniques that exploit visual features such as edges and texture to unobtrusively communicate information. These can then be layered over existing techniques to further boost the capacity. Our experimental results show that there is potential to achieve an average goodput of about 22 kbps, significantly outperforming existing work while remaining flicker-free.

Simple, fast and lightweight SLAM algorithms are necessary in many embedded robotic systems which soon will be used in houses and offices in order to do various service tasks. In this paper the Orthogonal SLAM algorithm is presented as an answer to this need. In continuation of our previous work, the algorithm is extended to generate 3D maps and empirically validated by mapping the long corridor of our lab with the accuracy comparable with hand measured ground truth. The main contribution resides in the idea of reducing the complexity by using orthogonality constraint in indoor environments. This is done by mapping only planes that are parallel or perpendicular to each other which represent the main structure of most indoor environments. Having this assumption, we use an inclined sensor setup (fixed 2D SICK laser range finders) to generate 3D orthogonal maps. The algorithm is extremely fast since in each step it just processes one line of laser measurements.

This paper presents an experimental evaluation of a plane extraction method using various line extraction algorithms. Four different algorithms are chosen, which are well known in mobile robotics and computer vision. Experiments are performed on two sets of 25 range images either obtained by simulation or acquired by a proprietary 3D laser scanner. The performance of the range image segmentation is measured in terms of an average segment classification ratio. Moreover, the speed of the method is measured to conclude on the suitability for service robot applications.

Today, lightweight SLAM algorithms are needed in many embedded robotic systems. In this paper the orthogonal SLAM (OrthoSLAM ) algorithm is presented and empirically validated. The algorithm has constant time complexity in the state estimation and is capable to run real-time. The main contribution resides in the idea of reducing the complexity by means of an assumption on the environment. This is done by mapping only lines that are parallel or perpendicular to each other which represent the main structure of most indoor environments. The combination of this assumption with a Kalman filter and a relative map approach is able to map our laboratory hallway with the size of 80 m times 50 m and a trajectory of more than 500 m. The precision of the resulting map is similar to the measurements done by hand which are used as the ground-truth

In this paper, the independence relative map algorithm is presented. The algorithm aims to achieve the independence of relative map states. We show that using dependent relative quantities from the same observation creates a bias to the state covariance matrix, leading to an inaccurate and inconsistent relative map. Having independent map states improves the map consistency. Two case studies are presented in which we apply the proposed algorithm together with two popular relative map methods. Experimental results on simulated data show that the integrated algorithms outperform the original methods in term of map consistency and algorithm speed

In [5], a version of Relative Map Filter (RMF) is proposed to solve the simultaneous localization and map building (SLAM) problem. In the RMF, the map states contain only quantities invariant under shift and rotation. The estimation of the map states and their correlations is carried out in an optimal way using the Kalman filter. However, the dependency among the map states is not taken into account, thus the resulting map states are inconsistent. This paper presents two methods to enforce the consistency of the relative map states. The idea is to maintain a geometrically consistent map by solving a set of constraints between the map states. Experimental results obtained by using the proposed methods on real platform data show better performance than those deduced from the original RMF.

This paper presents an experimental evaluation of different line extraction algorithms on 2D laser scans for indoor environment. Six popular algorithms in mobile robotics and computer vision are selected and tested. Experiments are performed on 100 real data scans collected in an office environment with a map size of 80m /spl times/ 50m. Several comparison criteria are proposed and discussed to highlight the advantages and drawbacks of each algorithm, including speed, complexity, correctness and precision. The results of the algorithms are compared with the ground truth using standard statistical methods.