Abstract

While traditional stable matching algorithms, such as the Gale-Shapley algorithm, prioritize stability, they may fall short of achieving equitable outcomes among participants. We study the role of emph{Nash social welfare} (NSW) as a fairness objective in the classic emph{stable marriage problem}. We develop texttt{SNSW-Alg} that finds a stable matching that maximizes Nash social welfare under rank-induced utilities in $tilde{mathcal{O}}(n^4)$ time, where $n$ is the number of men or women. We demonstrate that texttt{SNSW-Alg} balances equity while preserving stability. We empirically evaluate our methods across diverse preference distributions, demonstrating significant gains in fairness without substantial losses in other key measures such as regret, egalitarian criterion, and sex equality. Our findings suggest that the stable matching produced by texttt{SNSW-Alg} is statistically Pareto-undominated by stable matchings based on other fairness measures - regret, egalitarian, and sex equality. This study offers compelling insights into designing fair stable matching.

Abstract

The growing significance of data-intensive technologies has positioned data as a critical geopolitical asset, prompting the emergence of data diplomacy as an increasingly important domain in contemporary foreign policy. This article examines the strategic relevance of data diplomacy in India's foreign policy through a qualitative analysis of policy documents, legislative frameworks and international engagements in the digital domain. It examines data diplomacy as a mutually re-enforcing process in which domestic data governance and international digital strategy are interrelated. The article analyses how domestic regulatory choices shape India's global negotiating positions while geopolitical, technological and economic imperatives simultaneously influence the evolution of domestic data governance. By locating areas of cooperation and contestation, the article demonstrates how data diplomacy is emerging as a crucial mechanism in India's digital strategic outlook through which it seeks to reconcile multiple objectives such as developmental priorities, digital sovereignty, economic competitiveness and strategic autonomy within an evolving global digital order.

Abstract

As data flows across borders become increasingly critical, governments are faced with the challenge
of governing these flows while maximizing their economic and strategic benefits. This article applies
dependency theory (Dos Santos, 1970, The American Economic Review, 60, 231-236; Wallerstein,
1974, The modern world-system i: Capitalist agriculture and the origins of the European world-economy in the
sixteenth century, University of California Press) as an analytical framework, examining how data
flows and governance systems create and reinforce dependent relationships between countries.
This framework proposes three categories of actors in the global economy: the core, periphery and
semi-periphery. Originally developed to analyse colonial and post-colonial economic relationships, this
theory can be used to examine how contemporary data governance may perpetuate global inequali-
ties through new mechanisms. In this article, we extend the dependency theory framework to the
global data economy, by analysing the evolving characteristics of the participating actors. Countries that
control key technological infrastructure, set global standards and capture the majority of economic
value from data flows can be characterized as the 'core'. Countries which serve as data sources, with
limited domestic technological capabilities and minimal value capture from their own data generation,
would be considered the 'periphery'. On the other hand, countries that are able to maintain a strate-
gic and dynamic trade and regulatory positioning, can exhibit selective participation and have growing
indigenous technological capabilities would form the 'semi-periphery'. Through this article, we seek to
apply dependency theory to analyse the global data economy and to situate India within it, highlighting
its characteristics as an actor in the semi-periphery.

Abstract

Identity-aware SIEM alerting compares activity with expected role context such as locations, managed devices, schedules, and entitlements. Legitimate organizational change makes that context stale, while unconstrained adaptation can learn attacker-controlled context as normal. We study who may mutate this persistent security state. CHANGEGUARD treats statistical drift and an exact governance record as evidence for a candidate, rather than permission to update the live baseline. Their intersection creates a role-scoped shadow; promotion additionally requires measured alert benefit, evidence from several principals, an attack-evidence veto, and bounded regression on frozen known attack families. On 467030 controlled synthetic windows from 12 fresh held-out profiles and 3 replays, CHANGEGUARD reduces post-change false-positive rate from 6.44% to 2.55%. Its absolute later-exploit recall is 48.27%, retaining 98.59% of the detections available to the frozen detector; persistent drift reaches a similar 2.59% false-positive rate but only 25.62% recall. Under mixed governance faults, drift-plus-record makes 26 over-broad mutations and CHANGEGUARD makes none. Applying the same frozen model and controller to a separately implemented 1356831-window process generator reproduces both findings. White-box stress also identifies the information boundary: coordinated attacks that reproduce every trusted input can pass. These results establish controlled mechanism behavior rather than production-SOC performance.

Abstract

An identity-related security alert can retain a low risk score even when the records needed to justify deprioritizing it are missing. CAW-Triage separates risk scoring, bounded waiting for telemetry, and permission to assign low priority. A versioned contract specifies required observations, completeness, and score/support conditions. Predicted recovery cannot substitute for observation. We evaluate 12,480 synthetic alert windows from twelve parameterizations of one scenario generator, with twenty degraded-telemetry replays per profile. At identical action counts, unsafe downgrade falls from 4.85% for a 15-minute ranker to 0.97%, a 3.88 percentage-point reduction (paired crossed-bootstrap 95% interval 3.08-4.68). High and review together account for 47.21% of alerts; review alone accounts for 26.89%. Fixed-window contracts achieve lower unsafe downgrade but higher load; recovery sensitivity exposes the additional delay caused by optimistic health estimates. A learned selector also achieves lower label-based error, but violates the declared contract on 23.0% of its low assignments. The results support conditional enforcement with measured trade-offs, not universal predictive superiority or production safety. Independent policy validation and enterprise telemetry remain necessary.

Abstract

Traditional public-key cryptographic schemes are primarily based on the computational hardness of problems such as the "Integer Factorization Problem (IFP)" and the "Elliptic Curve Discrete Logarithm Problem (ECDLP)". However, with the advent of large-scale quantum computing, these problems can be solved efficiently, thereby compromising the security of conventional cryptographic systems. Lattice-based cryptography has emerged as a promising solution due to its strong resistance to quantum attacks. In this work, we propose a post-quantum secure access control scheme based on lattice cryptography for IoMT-enabled healthcare environments. The scheme ensures that only authenticated and integrity-protected medical data are recorded in a private blockchain maintained collaboratively by multiple healthcare institutions. The proposed framework not only achieves strong security against both classical and quantum attacks but also maintains computational efficiency when compared to existing literature solutions.

Abstract

QMUL-SurvFace is the only large-scale surveillance face recognition benchmark with natively low-resolution faces; alternative benchmarks use high-resolution images that are downsampled only during evaluation. The median face in QMUL-SurvFace is just 19 pixels wide, meaning that individual facial features may occupy only one or two pixels.

Synthetic data offers an attractive way to increase identity coverage at this resolution. However, faces generated by current synthetic data pipelines are typically clean portraits, and mixing them with surveillance data can increase the domain gap rather than reduce it. We show that synthetic data is beneficial when the degradation pipeline is calibrated to the target dataset's per-channel pixel statistics. In other words, simply applying a generic augmentation pipeline designed to represent "surveillance" is insufficient; the specific motion blur, color cast, and JPEG characteristics of the target dataset must be matched.

The calibrated degradation pipeline transfers successfully between two unrelated synthetic identity sources, Arc2Face and DigiFace-1M, without requiring parameter retuning. A replication using CentreFace with the same backbone also reproduces the observed improvement, indicating that the primary source of the gain is the data rather than the choice of margin loss.

Using an IR-100 backbone and training with real and synthetic data jointly from the first epoch, we achieve 45.73% TAR at FAR = 0.1%, representing an 18.66 percentage-point improvement over the real-only baseline. The same model also outperforms the real-only baseline on the QMUL-SurvFace open-set identification protocol, improving performance by 4.16 percentage points at FPIR = 10% and nearly doubling the rate at FPIR = 1%.

We additionally observe a loss-data interaction in single-seed experiments: calibrated degradation provides a 3.3 percentage-point improvement with ArcFace, while it results in a 3.0 percentage-point decrease with AdaFace. This is consistent with AdaFace's quality-adaptive margin, which down-weights low-norm samples during training. However, multi-seed reruns place both effects within the observed seed variance. Therefore, we treat this interaction as a smaller, dataset-dependent observation rather than a definitive mechanism claim.

Abstract

Sharing genuine multipartite entanglement by considering the collective use of copies of biseparable states,
which are entangled across all bipartitions but lack genuine multipartite entanglement at the single-copy level,
plays a central role in several quantum information processing protocols, and has been referred to as genuine
multipartite entanglement activation. We present a protocol for three-qutrit systems showing that two copies of
rank-two biseparable states, entangled across every bipartition, are sufficient to generate a genuinely multipartite
entangled state with nonzero probability. This contrasts with the three-qubit scenario where many copies of
biseparable states might be required for sharing genuine multipartite entanglement. We subsequently generalize
our protocols to the case of an arbitrary number of parties. Interestingly, the proposed construction naturally leads
to the activation of genuinely nonlocal correlations, yielding a result that is stronger than genuine multipartite
entanglement activation alone.

Abstract

For applications like flood forecasting, water-resource management, and understanding the temporal structure of the Indian monsoon, accurate sub-daily (hourly) near-real-time (NRT) precipitation observations are critical. These NRT estimates from Satellite Precipitation Products (SPPs) fill the gap left by sparse gauge networks but introduce a systematic diurnal phase shift and spatial displacement errors. Crucially, studies regarding the performance of these NRT SPPs in representing diurnal precipitation characteristics in the geographical region of India are unavailable. This paper benchmarks three operational NRT products (IMERG-E, GSMaP-NRT, and CMORPH-RT) against the gauge-corrected GSMaP-ISRO reference using a conceptually increasing suite of benchmarking metrics: Harmonic Vector Dials, Earth Mover's Distance, and Unbalanced Sinkhorn Divergence (USD) across Hyderabad during summer monsoon season of 2025. IMERG-E and GSMaP-NRT showed comparable spatial-temporal errors with USD values of 31.59 and 31.05, respectively, and consistently outperformed CMORPH-RT, establishing them as the most reliable NRT products over Hyderabad study region.

Abstract

Recent advances in Large Language Models (LLMs) and Vision Language Models (VLMs) have enabled their adaptation to remote-sensing tasks, including Remote Sensing Visual Question Answering (RSVQA). However, most existing approaches rely primarily on RGB imagery, often underutilising rich information contained in non-visible spectral bands. Moreover, Language Models (LMs), which form the backbone of such pipelines, are unreliable at arithmetic and logical reasoning over pre-extracted area values and are prone to hallucination when processing large numerical contexts, offering limited interpretability. In this work, we propose a database-grounded pipeline in which the spatial-spectral context extracted from satellite imagery is stored in a lightweight relational database. A small language model (Qwen2.5-7B) is then used to generate SQLite queries that retrieve and process the required information to answer the user's question. We compare the proposed approach with two prompt-based baselines: (i) a full-context approach that injects all extracted context into the prompt, and (ii) a selective-context approach that injects only the subset of context relevant to the query. Experimental results show that the proposed pipeline consistently outperforms both baselines on arithmetic and logical reasoning over pre-extracted numerical information, as well as on location-based reasoning tasks.