Rebati R. Gaire

Active learning (AL) offers a promising strategy for reducing annotation costs by selectively querying informative samples. However, its deployment on edge devices remains fundamentally limited. In such resource-constrained environments, models must be highly compact to meet strict compute, memory, and energy budgets. These lightweight models, though efficient, suffer from limited representational capacity and are ill-equipped to support existing AL methods, which assume access to high-capacity networks capable of modeling uncertainty or learning expressive acquisition functions. To address this, we introduce FDAL, a unified framework that couples task-aware AL with feature-distilled training to enable efficient and accurate learning on resource-limited devices. A task-aware sampler network, trained adversarially alongside a lightweight task model, exploits refined features from feature distillation to prioritize informative unlabeled instances for annotation. This joint optimization strategy ensures tight coupling between task utility and sampling efficacy. Extensive experiments on SVHN, CIFAR-10, and CIFAR-100 demonstrate that FDAL consistently outperforms state-of-the-art AL methods, achieving competitive accuracy with significantly fewer labels under limited compute and annotation budgets. Notably, FDAL achieves 78.5% accuracy on CIFAR-10 with only 30% labeled data, matching the fully supervised baseline of 78.38%. The code is made publicly available at https://github.com/rrgaire/FDAL for reproducibility and future research.

Deep neural networks (DNNs) have achieved remarkable success across various domains, yet their rigid, static computation graphs lead to significant inefficiencies in real-world deployment. Standard architectures allocate equal computational resources to all inputs, disregarding their inherent complexity, which results in unnecessary computation for simple samples and suboptimal processing for complex ones. To address this, we propose the Complexity-Aware Routing Network (CARN), a novel framework that dynamically adjusts computational pathways based on input complexity. CARN integrates a self-supervised complexity estimation module that quantifies input difficulty using confidence, entropy, and computational cost, guiding a neural network-based routing mechanism to optimally assign task modules. The model is trained using a routing loss function that balances assignment accuracy and computational efficiency, mitigating expert starvation while preserving specialization. Extensive experiments on CIFAR-10, CIFAR-100, and Tiny-ImageNet demonstrate that CARN achieves up to 4x reduction in computational cost and over 10x reduction in parameter movement while maintaining high accuracy compared to state-of-the-art static models. The code and pre-trained models are made available at https://github.com/rrgaire/CARN for reproducibility and further research.

Artificial-intelligence-powered edge devices are inspiring interest in always-on, intelligent, and self-powered visual perception systems. Due to the high energy cost of converting raw data and the limited computing and energy resources available, designing energy-efficient and low bandwidth CMOS vision sensors is vital as these emerging systems require continuous sensing and instant processing. This paper proposes a low-power integrated sensing and computing engine, namely APRIS, including a novel software/hardware co-design technique. This method provides a highly parallel analog multiplication and accumulation-in-pixel scheme, which realizes low-precision quantized weight neural networks to mitigate the overhead of analog-to-digital converters and analog buffers. Moreover, in order to reduce the size and power consumption, we propose the implementation of an approximate ADC in the readout circuit. Our system utilizes eight memory banks to increase computation parallelism, which has a dramatic effect on its speed and efficiency. Moreover, the proposed structure supports a zero-skipping scheme to reduce power consumption further. Our circuit-to-application co-simulation results demonstrate a comparable accuracy for our platform to the full-precision baseline on various object classification tasks while reaching an efficiency of 3.48 TOp/s/W.

This paper presents Encode, a novel technique that merges active learning, model compression, and knowledge distillation to optimize deep learning models. The method tackles issues such as generalization loss, resource intensity, and data redundancy that usually impede compressed models’ performance. It actively integrates valuable samples for labeling, thus enhancing the student model’s performance while economizing on labeled data and computational resources. Encode’s utility is empirically validated using SVHN and CIFAR-10 datasets, demonstrating improved model compactness, enhanced generalization, reduced computational complexity, and lessened labeling efforts. In our evaluations, applied to compressed versions of VGGll and AlexNet models, Encode consistently outperforms baselines even when trained with 60% of the total training samples. Thus, it establishes an effective framework for enhancing the accuracy and generalization capabilities of compressed models, which is especially beneficial in situations with limited resources and scarce labeled data.

We present our novel deep multi-task learning method for medical image segmentation. Existing multi-task methods demand ground truth annotations for both the primary and auxiliary tasks. Contrary to it, we propose to generate the pseudo-labels of an auxiliary task in an unsupervised manner. To generate the pseudo-labels, we leverage Histogram of Oriented Gradients (HOGs), one of the most widely used and powerful hand-crafted features for detection. Together with the ground truth semantic segmentation masks for the primary task and pseudo-labels for the auxiliary task, we learn the parameters of the deep network to minimize the loss of both the primary task and the auxiliary task jointly. We employed our method on two powerful and widely used semantic segmentation networks: UNet and U2Net to train in a multi-task setup. To validate our hypothesis, we performed experiments on two different medical image segmentation data sets. From the extensive quantitative and qualitative results, we observe that our method consistently improves the performance compared to the counter-part method. Moreover, our method is the winner of FetReg Endovis Sub-challenge on Semantic Segmentation organised in conjunction with MICCAI 2021. Code and implementation details are available at: https://github.com/thetna/medical_image_segmentation .

International benchmarking competitions have become fundamental for the comparative performance assessment of image analysis methods. However, little attention has been given to investigating what can be learnt from these competitions. Do they really generate scientific progress? What are common and successful participation strategies? What makes a solution superior to a competing method? To address this gap in the literature, we performed a multi-center study with all 80 competitions that were conducted in the scope of IEEE ISBI 2021 and MICCAI 2021. Statistical analyses performed based on comprehensive descriptions of the submitted algorithms linked to their rank as well as the underlying participation strategies revealed common characteristics of winning solutions. These typically include the use of multi-task learning (63%) and/or multi-stage pipelines (61%), and a focus on augmentation (100%), image preprocessing (97%), data curation (79%), and postprocessing (66%). The "typical" lead of a winning team is a computer scientist with a doctoral degree, five years of experience in biomedical image analysis, and four years of experience in deep learning. Two core general development strategies stood out for highly-ranked teams: the reflection of the metrics in the method design and the focus on analyzing and handling failure cases. According to the organizers, 43% of the winning algorithms exceeded the state of the art but only 11% completely solved the respective domain problem. The insights of our study could help researchers (1) improve algorithm development strategies when approaching new problems, and (2) focus on open research questions revealed by this work.

Fetoscopy laser photocoagulation is a widely adopted procedure for treating Twin-to-Twin Transfusion Syndrome (TTTS). The procedure involves photocoagulation pathological anastomoses to restore a physiological blood exchange among twins. The procedure is particularly challenging, from the surgeon’s side, due to the limited field of view, poor manoeuvrability of the fetoscope, poor visibility due to amniotic fluid turbidity, and variability in illumination. These challenges may lead to increased surgery time and incomplete ablation of pathological anastomoses, resulting in persistent TTTS. Computer-assisted intervention (CAI) can provide TTTS surgeons with decision support and context awareness by identifying key structures in the scene and expanding the fetoscopic field of view through video mosaicking. Research in this domain has been hampered by the lack of high-quality data to design, develop and test CAI algorithms. Through the Fetoscopic Placental Vessel Segmentation and Registration (FetReg2021) challenge, which was organized as part of the MICCAI2021 Endoscopic Vision (EndoVis) challenge, we released the first large-scale multi-center TTTS dataset for the development of generalized and robust semantic segmentation and video mosaicking algorithms with a focus on creating drift-free mosaics from long duration fetoscopy videos. For this challenge, we released a dataset of 2060 images, pixel-annotated for vessels, tool, fetus and background classes, from 18 in-vivo TTTS fetoscopy procedures and 18 short video clips of an average length of 411 frames for developing placental scene segmentation and frame registration for mosaicking techniques. Seven teams participated in this challenge and their model performance was assessed on an unseen test dataset of 658 pixel-annotated images from 6 fetoscopic procedures and 6 short clips. For the segmentation task, overall baseline performed was the top performing (aggregated mIoU of 0.6763) and was the best on the vessel class (mIoU of 0.5817) while team RREB was the best on the tool (mIoU of 0.6335) and fetus (mIoU of 0.5178) classes. For the registration task, overall the baseline performed better than team SANO with an overall mean 5-frame SSIM of 0.9348. Qualitatively, it was observed that team SANO performed better in planar scenarios, while baseline was better in non-planner scenarios. The detailed analysis showed that no single team outperformed on all 6 test fetoscopic videos. The challenge provided an opportunity to create generalized solutions for fetoscopic scene understanding and mosaicking. In this paper, we present the findings of the FetReg2021 challenge, alongside reporting a detailed literature review for CAI in TTTS fetoscopy. Through this challenge, its analysis and the release of multi-center fetoscopic data, we provide a benchmark for future research in this field.