Growing concerns surrounding AI safety and data privacy have driven the development of Machine Unlearning as a potential solution. However, current machine unlearning algorithms are designed to complement the offline training paradigm. The emergence of the Continual Learning (CL) paradigm promises incremental model updates, enabling models to learn new tasks sequentially. Naturally, some of those tasks may need to be unlearned to address safety or privacy concerns that might arise. We find that applying conventional unlearning algorithms in continual learning environments creates two critical problems: performance degradation on retained tasks and task relapse, where previously unlearned tasks resurface during subsequent learning. Furthermore, most unlearning algorithms require data to operate, which conflicts with CL's philosophy of discarding past data. A clear need arises for unlearning algorithms that are data-free and mindful of future learning. To that end, we propose UnCLe, an Unlearning framework for Continual Learning. UnCLe employs a hypernetwork that learns to generate task-specific network parameters, using task embeddings. Tasks are unlearned by aligning the corresponding generated network parameters with noise, without requiring any data. Empirical evaluations on several vision data sets demonstrate UnCLe's ability to sequentially perform multiple learning and unlearning operations with minimal disruption to previously acquired knowledge.

Continuous and reliable access to curated biological data repositories is indispensable for accelerating rigorous scientific inquiry and fostering reproducible research. Centralized repositories, though widely used, are vulnerable to single points of failure arising from cyberattacks, technical faults, natural disasters, or funding and political uncertainties. This can lead to widespread data unavailability, data loss, integrity compromises, and substantial delays in critical research, ultimately impeding scientific progress. Centralizing essential scientific resources in a single geopolitical or institutional hub is inherently dangerous, as any disruption can paralyze diverse ongoing research. The rapid acceleration of data generation, combined with an increasingly volatile global landscape, necessitates a critical re-evaluation of the sustainability of centralized models. Implementing federated and decentralized architectures presents a compelling and future-oriented pathway to substantially strengthen the resilience of scientific data infrastructures, thereby mitigating vulnerabilities and ensuring the long-term integrity of data. Here, we examine the structural limitations of centralized repositories, evaluate federated and decentralized models, and propose a hybrid framework for resilient, FAIR, and sustainable scientific data stewardship. Such an approach offers a significant reduction in exposure to governance instability, infrastructural fragility, and funding volatility, and also fosters fairness and global accessibility. The future of open science depends on integrating these complementary approaches to establish a globally distributed, economically sustainable, and institutionally robust infrastructure that safeguards scientific data as a public good, further ensuring continued accessibility, interoperability, and preservation for generations to come.

Artificial intelligence is poised to augment dermatological care by enabling scalable image-based diagnostics. Yet, the development of robust and equitable models remains hindered by datasets that fail to capture the clinical and demographic complexity of real-world practice. This complexity stems from region-specific disease distributions, wide variation in skin tones, and the underrepresentation of outpatient scenarios from non-Western populations. We introduce DermaCon-IN, a prospectively curated dermatology dataset comprising over 5,450 clinical images from approximately 3,000 patients across outpatient clinics in South India. Each image is annotated by board-certified dermatologists with over 240 distinct diagnoses, structured under a hierarchical, etiology-based taxonomy adapted from Rook's classification. The dataset captures a wide spectrum of dermatologic conditions and tonal variation commonly seen in Indian outpatient care. We benchmark a range of architectures including convolutional models (ResNet, DenseNet, EfficientNet), transformer-based models (ViT, MaxViT, Swin), and Concept Bottleneck Models to establish baseline performance and explore how anatomical and concept-level cues may be integrated. These results are intended to guide future efforts toward interpretable and clinically realistic models. DermaCon-IN provides a scalable and representative foundation for advancing dermatology AI in real-world settings.

Accurate uncertainty quantification remains a key challenge for standard LLMs, prompting the adoption of Bayesian and ensemble-based methods. However, such methods typically necessitate computationally expensive sampling, involving multiple forward passes to effectively estimate predictive uncertainty. In this paper, we introduce a novel approach enabling efficient and effective uncertainty estimation in LLMs without sacrificing performance. Specifically, we distill uncertainty-aware teacher models - originally requiring multiple forward passes - into compact student models sharing the same architecture but fine-tuned using Low-Rank Adaptation (LoRA). We compare two distinct distillation strategies: one in which the student employs traditional softmax-based outputs, and another in which the student leverages Dirichlet-distributed outputs to explicitly model epistemic uncertainty via evidential learning. Empirical evaluations on classification datasets demonstrate that such students can achieve comparable or superior predictive and uncertainty quantification performance relative to their teacher models, while critically requiring only a single forward pass. To our knowledge, this is the first demonstration that immediate and robust uncertainty quantification can be achieved in LLMs through evidential distillation.

In this work, we study the holographic entanglement entropy (HEE) and holographic complexity (HC) for three-dimensional dyonic quantum black holes, incorporating corrections arising from bulk quantum fields in the setup of double holography. We investigate the holographic entanglement entropy through the holographic Ryu-Takayanagi (RT) prescription and the island prescription. Using RT extremization, we evaluate HEE for connected and disconnected (island) surfaces and show islands emerge when RT surfaces intersect the brane; entanglement entropy grows with subregion size and ultimately saturates for quantum black holes as well as dressed defects. For complexity, we analyze both CV (perturbative) and CA (exact, all-orders) prescriptions: the leading quantum corrections feature universal behavior and the late-time growth can be expressed in thermodynamic variables, obeying generalized Lloyd-type bounds. In contrast, quantum dressed defects exhibit vanishing late-time growth. The CA prescription proves to be more tractable nonperturbatively and yields a thermodynamic interpretation of complexity growth.

The ability to continuously learn and adapt itself to new tasks, without losing grasp of already acquired knowledge is a hallmark of biological learning systems, which current deep learning systems fall short of. In this work, we present a novel methodology for continual learning called MERLIN: Meta-Consolidation for Continual Learning. We assume that weights of a neural network $\boldsymbol \psi$, for solving task $\boldsymbol t$, come from a meta-distribution $p(\boldsymbol{\psi|t})$. This meta-distribution is learned and consolidated incrementally. We operate in the challenging online continual learning setting, where a data point is seen by the model only once. Our experiments with continual learning benchmarks of MNIST, CIFAR-10, CIFAR-100 and Mini-ImageNet datasets show consistent improvement over five baselines, including a recent state-of-the-art, corroborating the promise of MERLIN.

This paper introduces an adaptive convolutional neural network (CNN) architecture capable of automating various topology optimization (TO) problems with diverse underlying physics. The proposed architecture has an encoder-decoder-type structure with dense layers added at the bottleneck region to capture complex geometrical features. The network is trained using datasets obtained by the problem-specific open-source TO codes. Tensorflow and Keras are the main libraries employed to develop and to train the model. Effectiveness and robustness of the proposed adaptive CNN model are demonstrated through its performance in compliance minimization problems involving constant and design-dependent loads and in addressing bulk modulus optimization. Once trained, the model takes user's input of the volume fraction as an image and instantly generates an output image of optimized design. The proposed CNN produces high-quality results resembling those obtained via open-source TO codes with negligible performance and volume fraction errors. The paper includes complete associated Python code (Appendix A) for the proposed CNN architecture and explains each part of the code to facilitate reproducibility and ease of learning.

We experimentally investigate the wake dynamics of a square cylinder rising through quiescent water over a range of Froude numbers ($\mathrm{Fr}$). Time-resolved Particle Image Velocimetry provides velocity and vorticity fields that enable pressure reconstruction and vortex characterization. Diagnostics based on swirl strength ($\lambda_{ci}$), the Okubo-Weiss parameter ($W$), and a shear-vortex interaction measure ($\Lambda$) reveal that the wake is governed by a persistent pair of counter-rotating vortices rather than by periodic shedding. Circulation exhibits a two-regime dependence on $\mathrm{Fr}$, with a sharp increase below $\mathrm{Fr}\approx 1$ and saturation above this threshold, mirroring entrainment force scaling reported previously. While vortex area remains nearly constant, swirl strength and negative-$W$ regions expand with $\mathrm{Fr}$, indicating that entrainment enhancement arises from intensified rotation rather than an enlarged vortex footprint. These findings provide new physical insight into vortex-free-surface interactions and enrich the understanding of entrainment mechanisms in unsteady wakes, with implications for multiphase flows and the hydrodynamic design of naval and offshore structures.

The increasing demand for high data rates and seamless connectivity in wireless systems has sparked significant interest in reconfigurable intelligent surfaces (RIS) and artificial intelligence-based wireless applications. RIS typically comprises passive reflective antenna elements that control the wireless propagation environment by adequately tuning the phase of the reflective elements. The allocation of RIS elements to multipleuser equipment (UEs) is crucial for efficiently utilizing RIS. In this work, we formulate a joint optimization problem that optimizes the RIS phase configuration and resource allocation under an $\alpha$-fair scheduling framework and propose an efficient way of allocating RIS elements. Conventional iterative optimization methods, however, suffer from exponentially increasing computational complexity as the number of RIS elements increases and also complicate the generation of training labels for supervised learning. To overcome these challenges, we propose a five-layer fully connected neural network (FNN) combined with a preprocessing technique to significantly reduce input dimensionality, lower computational complexity, and enhance scalability. The simulation results show that our proposed NN-based solution reduces computational overhead while significantly improving system throughput by 6.8% compared to existing RIS element allocation schemes. Furthermore, the proposed system achieves better performance while reducing computational complexity, making it significantly more scalable than the iterative optimization algorithms.

Bibliometric measures, such as total citations and h-index, have become a cornerstone for evaluating academic performance; however, these traditional metrics, being non-weighted, inadequately capture the nuances of individual contributions. To address this constraint, we developed GScholarLens, an open-access browser extension that integrates seamlessly with Google Scholar to enable detailed bibliometric analysis. GScholarLens categorizes publications by authorship roles, adjusts citation weightings accordingly, and introduces Scholar h-index, Sh-index, an authorship-contribution normalized h-index. This tool proportionally weights citations based on authorship position using heuristic percentages, i.e., corresponding 100 percent, first 90 percent, second 50 percent, co-authors in publications with less than six authors 25 percent, and co-authors with more than six authors 10 percent. Currently, there is no empirical data available for author-contribution weights, however, this proof-of-concept framework can easily adapt more precise author-contribution weightage data decided by authors at the time of manuscript submission along with CRediT, which journals and publishers can mandate. Furthermore, this tool incorporates retraction detection by mapping data from retraction databases into the Google Scholar interface. By aligning bibliometric evaluation more closely with actual scholarly contribution, GScholarLens presents a better open-access framework for academic recognition, particularly within interdisciplinary and highly collaborative research environments. This tool is freely accessible at this https URL.

The newly discovered Galactic transient MAXI J1744-294 went into its first X-ray outburst in 2025. We study the spectral properties of this source in the 2-10 keV energy band during this outburst using X-ray data from the XRISM satellite for both of its Resolve and Xtend instruments, taken on March 03, 2025. High-resolution spectroscopy has revealed, for the first time, complex iron line features in this source, corresponding to distinct components of Fe XXV emission and Fe XXVI absorption lines. Such a detailed structure has not been reported in other low-mass X-ray binaries to date, prior to the XRISM era. Our analysis shows that the line complexes arise from two highly ionized plasmas with ionization rate ~ 1000 erg-cm/s with distinct turbulent velocities: one broad (~2513 km/s) from hot gas at the inner accretion disk and one narrow (~153 km/s) scattered by nearby photoionized gas. These results offer new insight into the reprocessing of continuum in stratified media, either in the accretion disk or winds, or both, for XRBs in the soft state. The data are well described by models with spin, mass of the black hole, and accretion disk inclination 0.63-0.70, 5.7-10.1 Solar masses, and 19-24 degrees. The fitted spectral model parameters suggest that the source is in the soft spectral state. The source is situated in a crowded field near the Galactic center, resulting in a large hydrogen column density.

In class-incremental learning, the model is expected to learn new classes continually while maintaining knowledge on previous classes. The challenge here lies in preserving the model's ability to effectively represent prior classes in the feature space, while adapting it to represent incoming new classes. We propose two distillation-based objectives for class incremental learning that leverage the structure of the feature space to maintain accuracy on previous classes, as well as enable learning the new classes. In our first objective, termed cross-space clustering (CSC), we propose to use the feature space structure of the previous model to characterize directions of optimization that maximally preserve the class: directions that all instances of a specific class should collectively optimize towards, and those that they should collectively optimize away from. Apart from minimizing forgetting, this indirectly encourages the model to cluster all instances of a class in the current feature space, and gives rise to a sense of herd-immunity, allowing all samples of a class to jointly combat the model from forgetting the class. Our second objective termed controlled transfer (CT) tackles incremental learning from an understudied perspective of inter-class transfer. CT explicitly approximates and conditions the current model on the semantic similarities between incrementally arriving classes and prior classes. This allows the model to learn classes in such a way that it maximizes positive forward transfer from similar prior classes, thus increasing plasticity, and minimizes negative backward transfer on dissimilar prior classes, whereby strengthening stability. We perform extensive experiments on two benchmark datasets, adding our method (CSCCT) on top of three prominent class-incremental learning methods. We observe consistent performance improvement on a variety of experimental settings.

We study the viability of light thermal dark matter (DM) in sub-GeV mass range in view of the stringent new DAMIC-M limits on DM-electron scattering. Considering a Dirac fermion singlet DM charged under a new Abelian gauge symmetry $U(1)$, we outline two possibilities: (i) family non-universal $U(1)$ gauge coupling with resonantly enhanced DM annihilation into standard model (SM) fermions and (ii) family universal dark $U(1)$ gauge symmetry where relic is set by DM annihilation into light gauge bosons. As an illustrative example of the first class of models, we consider a gauged $L_\mu-L_\tau$ extension of the SM having interesting detection prospects at several experiments. While both of these class of models lead to observed DM relic and consistency with DAMIC-M together with other experimental limits, the second class of models also lead to strong DM self-interactions, potentially solving the small-scale structure issues of cold dark matter. While a vast part of the parameter space in both the models is already ruled out, the current allowed region of parameter space can be further probed at ongoing or future experiments keeping the models testable.

In recent years, a lot of technological advances in computer science have aided software programmers to create innovative and real-time user-friendly software. With the creation of the software and the urging interest of people to learn to write software, there is a large collection of source codes that can be found on the web, also known as Big Code, which can be used as a source of data for driving the machine learning applications tending to solve certain software engineering problems. In this paper, we present COFO, a dataset consisting of 809 classes/problems with a total of 369K source codes written in C, C++, Java, and Python programming languages, along with other metadata such as code tags, problem specification, and input-output specifications. COFO has been scraped from the openly available Codeforces website using a selenium-beautifulsoup-python based scraper. We envision that this dataset can be useful for solving machine learning-based problems like program classification/recognition, tagging, predicting program properties, and code comprehension.