arXiv subjects

Mahesh Chandran

Publications and source records attributed to Mahesh Chandran.

14 recordsLinked to original sources

arXiv2026

Learning Long Range Spatio-Temporal Representations over Continuous Time Dynamic Graphs with State Space Models

Continuous-time dynamic graphs (CTDGs) provide a richer framework to capture fine-grained temporal patterns in evolving relational data. Long-range information propagation is a key challenge while learning representations, wherein it is important to retain and update information over long temporal horizons. Existing approaches restrict models to capture one-hop or local temporal neighborhoods and fail to capture multi-hop or global structural patterns. To mitigate this, we derive a parameter-efficient state-space modeling framework for continuous-time dynamic graphs (CTDG-SSM) from first principles. We first introduce continuous-time Topology-Aware higher order polynomial projection operator (CTT-HiPPO), a novel memory-based reformulation of HiPPO to jointly encode temporal dynamics and graph structure. The solution from CTT-HiPPO is obtained by projecting the classical HiPPO solution through a polynomial of the Laplacian matrix, yielding topology-aware memory updates that admit an equivalent state-space formulation for CTDGs (CTDG-SSM). Then a computationally efficient discrete formulation is obtained using the zero-order hold approach for model implementation. Across benchmarks on dynamic link prediction, dynamic node classification, and sequence classification, CTDG-SSM achieves state-of-the-art performance. Notably, it achieves large performance gains on datasets that require long range temporal (LRT) and spatial reasoning.

cs.LG↗

arXiv2026

HoT-SSM:Higher-order Temporal Knowledge Graph Reasoning with State Space Models for Health Care

Medical knowledge graphs (MKGs) infused with clinical knowledge have been increasingly used to model electronic health records (EHRs) to support interpretable predictions in healthcare domain. However, existing MKG-based approaches are limited in capturing pairwise relations between clinical concepts (e.g., conditions, procedures, and medications), and restricts their ability to model higher-order interactions among co-occurring or semantically related concepts. In addition, most representation learning methods that leverage MKGs either collapse temporal information across visits or lack an explicit mechanism for modeling long-range temporal dependencies, which is critical for clinical tasks such as mortality prediction. To mitigate these limitations, we propose HoT-SSM, a parameter efficient and higher-order temporal graph reasoning with state space models. For each visit, HoT-SSM constructs hypergraphs by grouping semantically related clinical concepts into hyperedges using domain knowledge, thereby preserving visit-level clinical context. Further, to model the temporal dynamics while learning the representations, we introduce a novel dynamic hypergraph-based state space model that explicitly captures patients latent state evolution over time while preserving long-range information. The learned representations are used for downstream clinical prediction and reasoning. Experiments on MIMIC-III and MIMIC-IV datasets shows significant performance improvement over the current state-of-the-art models, demonstrating the effectiveness of jointly modeling higher-order clinical interactions and long-range temporal dependencies.

cs.LG↗

arXiv2026

Self-Adaptive Graph Mixture of Models

Graph Neural Networks (GNNs) have emerged as powerful tools for learning over graph-structured data, yet recent studies have shown that their performance gains are beginning to plateau. In many cases, well-established models such as GCN and GAT, when appropriately tuned, can match or even exceed the performance of more complex, state-of-the-art architectures. This trend highlights a key limitation in the current landscape: the difficulty of selecting the most suitable model for a given graph task or dataset. To address this, we propose Self-Adaptive Graph Mixture of Models (SAGMM), a modular and practical framework that learns to automatically select and combine the most appropriate GNN models from a diverse pool of architectures. Unlike prior mixture-of-experts approaches that rely on variations of a single base model, SAGMM leverages architectural diversity and a topology-aware attention gating mechanism to adaptively assign experts to each node based on the structure of the input graph. To improve efficiency, SAGMM includes a pruning mechanism that reduces the number of active experts during training and inference without compromising performance. We also explore a training-efficient variant in which expert models are pretrained and frozen, and only the gating and task-specific layers are trained. We evaluate SAGMM on 16 benchmark datasets covering node classification, graph classification, regression, and link prediction tasks, and demonstrate that it consistently outperforms or matches leading GNN baselines and prior mixture-based methods, offering a robust and adaptive solution for real-world graph learning.

cs.LG↗

arXiv2026

PLGC: Pseudo-Labeled Graph Condensation

Large graph datasets make training graph neural networks (GNNs) computationally costly. Graph condensation methods address this by generating small synthetic graphs that approximate the original data. However, existing approaches rely on clean, supervised labels, which limits their reliability when labels are scarce, noisy, or inconsistent. We propose Pseudo-Labeled Graph Condensation (PLGC), a self-supervised framework that constructs latent pseudo-labels from node embeddings and optimizes condensed graphs to match the original graph's structural and feature statistics -- without requiring ground-truth labels. PLGC offers three key contributions: (1) A diagnosis of why supervised condensation fails under label noise and distribution shift. (2) A label-free condensation method that jointly learns latent prototypes and node assignments. (3) Theoretical guarantees showing that pseudo-labels preserve latent structural statistics of the original graph and ensure accurate embedding alignment. Empirically, across node classification and link prediction tasks, PLGC achieves competitive performance with state-of-the-art supervised condensation methods on clean datasets and exhibits substantial robustness under label noise, often outperforming all baselines by a significant margin. Our findings highlight the practical and theoretical advantages of self-supervised graph condensation in noisy or weakly-labeled environments.

cs.LG↗

arXiv2025

Can LLMs Help You at Work? A Sandbox for Evaluating LLM Agents in Enterprise Environments

Enterprise systems are crucial for enhancing productivity and decision-making among employees and customers. Integrating LLM based systems into enterprise systems enables intelligent automation, personalized experiences, and efficient information retrieval, driving operational efficiency and strategic growth. However, developing and evaluating such systems is challenging due to the inherent complexity of enterprise environments, where data is fragmented across multiple sources and governed by sophisticated access controls. We present EnterpriseBench, a comprehensive benchmark that simulates enterprise settings, featuring 500 diverse tasks across software engineering, HR, finance, and administrative domains. Our benchmark uniquely captures key enterprise characteristics including data source fragmentation, access control hierarchies, and cross-functional workflows. Additionally, we provide a novel data generation pipeline that creates internally consistent enterprise tasks from organizational metadata. Experiments with state-of-the-art LLM agents demonstrate that even the most capable models achieve only 41.8% task completion, highlighting significant opportunities for improvement in enterprise-focused AI systems.

cs.LG↗

arXiv2025

GnnXemplar: Exemplars to Explanations -- Natural Language Rules for Global GNN Interpretability

Graph Neural Networks (GNNs) are widely used for node classification, yet their opaque decision-making limits trust and adoption. While local explanations offer insights into individual predictions, global explanation methods, those that characterize an entire class, remain underdeveloped. Existing global explainers rely on motif discovery in small graphs, an approach that breaks down in large, real-world settings where subgraph repetition is rare, node attributes are high-dimensional, and predictions arise from complex structure-attribute interactions. We propose GnnXemplar, a novel global explainer inspired from Exemplar Theory from cognitive science. GnnXemplar identifies representative nodes in the GNN embedding space, exemplars, and explains predictions using natural language rules derived from their neighborhoods. Exemplar selection is framed as a coverage maximization problem over reverse k-nearest neighbors, for which we provide an efficient greedy approximation. To derive interpretable rules, we employ a self-refining prompt strategy using large language models (LLMs). Experiments across diverse benchmarks show that GnnXemplar significantly outperforms existing methods in fidelity, scalability, and human interpretability, as validated by a user study with 60 participants.

cs.LG↗

arXiv2006

Peak Effect in Superconductors: Absence of Phase Transition and Possibility of Jamming in Vortex Matter

The magnetic field $B$ dependence of the critical current $I_c$ for the vortex phase of a disordered superconductor is studied numerically at zero temperature. The $I_{c}(B)$ increases rapidly near the upper critical field $B_{c2}$ similar to the peak effect (PE) phenomenon observed in many superconductors. The real space configuration across the PE changes continuously from a partially ordered domain (polycrystalline) state into an amorphous state. The topological defect density $n_{d}(B)\sim e^{αB^{k}}$ with $k>1$ for $B\geq 0.4B_{c2}$. There is no evidence of a phase transition in the vicinity of the PE suggesting that an order-disorder transition is not essential for the occurrence of the PE phenomenon. An alternative view is presented wherein the vortex system with high dislocation density undergoes jamming at the onset of the PE.

cond-mat.supr-con↗

arXiv2004

Metastability and uniqueness of vortex states at depinning

We present results from numerical simulations of transport of vortices in the zero-field cooled (ZFC) and the field-cooled (FC) state of a type-II superconductor. In the absence of an applied current $I$, we find that the FC state has a lower defect density than the ZFC state, and is stable against thermal cycling. On the other hand, by cycling $I$, surprisingly we find that the ZFC state is the stable state. The FC state is metastable as manifested by increasing $I$ to the depinning current $I_{c}$, in which case the FC state evolves into the ZFC state. We also find that all configurations acquire a unique defect density at the depinning transition independent of the history of the initial states.

cond-mat.supr-con↗

arXiv2004

Fractionalization of a flux quantum in a one-dimensional parallel Josephson junction array with alternating $π$ junctions

We study numerically and analytically the properties of a one-dimensional array of parallel Josephson junctions in which every {\em alternate} junction is a $π$ junction. In the ground state of the array, each cell contains spontaneous magnetic flux $Φ\leqΦ_{0}/2$ which shows {\em antiferromagnetic} ordering along the array. We find that an externally introduced $2π$-fluxon $Φ_{0}$ in such an array is unstable and fractionalizes into two $π$ fluxons of magnitude ${1/2}Φ_{0}$. We attribute this fractionalization to the degeneracy of the ground state of the array. The magnitude of the flux in the fractional fluxons can be controlled by changing the critical current of the $π$ junctions relative to the 0 junctions. In the presence of an external current, the fluxon lattice in the antiferromagnetic ground state can be depinned. We also observe a novel resonant structure in the $V$-$I$ characteristics above the depinning current due to the interaction between the fluxon lattice and the array.

cond-mat.supr-con↗

arXiv2004

Domain regime in two-dimensional disordered vortex matter

A detailed numerical study of the real space configuration of vortices in disordered superconductors using 2D London-Langevin model is presented. The magnetic field $B$ is varied between 0 and $B_{c2}$ for various pinning strengths $Δ$. For weak pinning, an inhomogeneous disordered vortex matter is observed, in which the topologically ordered vortex lattice survives in large domains. The majority of the dislocations in this state are confined to the grain boundaries/domain walls. Such quasi-ordered configurations are observed in the intermediate fields, and we refer it as the domain regime (DR). The DR is distinct from the low-field and the high-fields amorphous regimes which are characterized by a homogeneous distribution of defects over the entire system. Analysis of the real space configuration suggests domain wall roughening as a possible mechanism for the crossover from the DR to the high-field amorphous regime. The DR also shows a sharp crossover to the high temperature vortex liquid phase. The domain size distribution and the roughness exponent of the lattice in the DR are also calculated. The results are compared with some of the recent Bitter decoration experiments.

cond-mat.supr-con↗

arXiv2003

Dynamic transition in driven vortices across the peak effect in superconductors

We study the zero-temperature dynamic transition from the disordered flow to an ordered flow state in driven vortices in type-II superconductors. The transition current $I_{p}$ is marked by a sharp kink in the $V(I)$ characteristic with a concomitant large increase in the defect concentration. On increasing magnetic field $B$, the $I_{p}(B)$ follows the behaviour of the critical current $I_{c}(B)$. Specifically, in the peak effect regime $I_{p}(B)$ increases rapidly along with $I_{c}$. We also discuss the effect of varying disorder strength on $I_{p}$.

cond-mat.supr-con↗

arXiv2002

Implementation of parallel algorithms for 2D vortex dynamics simulation in type-II superconductors

This report discusses the implementation of two parallel algorithms on a distributed memory system for studying vortex dynamics in type-II superconductors. These algorithms are the same as that implemented for classical molecular dynamics simulation with short-range forces (Plimpton, J. Comp. Phys. Vol.117, 1 (1995)). The run time for parallel algorithm is tested on a system containing upto 4 processors and compared with that for vectorized algorithm on a single processor for system size ranging from 120 to 4800 vortices.

cond-mat.supr-con↗

arXiv2000

Magnetic relaxation in hard type-II superconductors

Magnetic relaxation in a type-II superconductor is simulated for a range of temperatures (T) in a simple model of 2D Josephson junction array (JJA) with finite screening. The high-T phase, that is characterised by a single time scale τ_α, crosses over to an intermediate phase at a lower temperature T_{cr} wherein a second time scale τ_β<<τ_α emerges. The relaxation in the time window set by τ_β follows power law which is attributed to self-organization of the magnetic flux during relaxation. Consequently, for T J_{c}) to sub-critical (J<J_{c}) state separated by an intermediate state with frozen dynamics is observed. Both τ_α and τ_β diverges at T_{sc}<T_{cr}, marking the transition into a state with true persistent current.

cond-mat.supr-con↗

arXiv2000

Irreversibility temperature from magnetic relaxation

Simulation of the magnetic relaxation in a model of hard superconductor reveals a new time scale below the irreversibility temperature. The relaxation in this new time scale, which appears in an intermediate time window, is a power law and is related to self organization of the magnetic flux around the critical current density J_c during the relaxation. Emergence of a new time scale which along with the time scale for long time relaxation due to thermally activated process dynamically identifies the irreversibility temperature.

cond-mat.supr-con↗