arXiv subjects

Shivanshu Dwivedi

Publications and source records attributed to Shivanshu Dwivedi.

4 recordsLinked to original sources

arXiv2026

Ghost Vectors: Soft-Deleted Embeddings Remain Reconstructible in HNSW Vector Databases

Retrieval-augmented generation (RAG) allows large language models to access external and private corpora for factual, domain-specific responses. Modern RAG pipelines use hierarchical navigable small world (HNSW) vector databases for efficient similarity search. When a user requests data deletion, the systems typically only mark the record as deleted, leaving the embedding on disk physically unchanged. This soft-delete operation raises compliance concerns under data-erasure and retention requirements such as GDPR Article 17 and HIPAA. Analysis on three HNSW implementations confirms that deleted vectors remain physically recoverable by accessing the raw index files at the storage layer, bypassing API access. Using the Vec2Text inversion model without domain-specific fine-tuning, we show this vulnerability on multiple real-world datasets and data modalities. On Wikipedia biographical living persons dataset (BLP), we successfully recover 25.5% of exact person names and 46.4% of geographic locations (ROUGE-L 0.185). Recovery reaches 100% for both patient age and gender markers (ROUGE-L 0.290) on highly structured, sensitive data (NIH Synthea dataset). On soft-deleted image embeddings, we show 100% tissue classification on histopathology patches (p=1.02e-07) and top-1 identity recovery reaches 99% on facial embeddings (p<0.01). This work introduces Epoch Key Rotation, which encrypts vectors and discards the key upon deletion. Epoch key rotation reduces observed PII recovery to 0% and completes in 2.5 ms for 500 deleted vectors (approximately 0.005 ms/record). Additionally, it generates an ECDSA-signed cryptographic proof as an auditable record of the deletion event.

cs.CR↗

arXiv2025

Comparative Analysis of Autonomous and Systematic Control Strategies for Hole-Doped Hubbard Clusters: Reinforcement Learning versus Physics-Guided Design

Engineering electron correlations in quantum dot arrays demands navigation of high-dimensional, non-convex parameter spaces where hole doping fundamentally alters the physics. We present a comparative study of two control paradigms for the one-hole, half-filled Hubbard model: (i) systematic physics-guided design and (ii) autonomous deep reinforcement learning with geometry-aware neural architectures. While systematic analysis reveals key design principles, such as field-induced localization for trapping the mobile hole, it becomes computationally intractable for optimization. We show that an autonomous RL agent, benchmarked across five 3D lattices from tetrahedron to FCC, achieves human-competitive accuracy (R^2 > 0.97) and 95.5 percent success on held-out tasks. The agent is 3-4 orders of magnitude more sample-efficient than grid search and outperforms other black-box optimization methods. Transfer learning yields 91 percent few-shot generalization to unseen geometries. This work establishes autonomous RL as a viable and highly efficient framework for rapid optimization and non-obvious strategy discovery in complex quantum systems.

cond-mat.str-el↗

arXiv2025

Microseismic Noise Mitigation with Machine Learning for Advanced LIGO

The unprecedented sensitivity of the Laser Interferometer Gravitational-Wave Observatory, which enables the detection of distant astrophysical sources, also renders the detectors highly susceptible to low-frequency ground motion. Persistent microseisms in the 0.1-0.3 Hz band couple into the instruments, degrade lock stability, and contribute substantially to detector downtime during observing runs. The multi-stage seismic isolation system has achieved remarkable success in mitigating such disturbances through active feedback control, yet residual platform motion remains a key factor limiting low-frequency sensitivity and duty cycle. Further reduction of this residual motion is therefore critical for improving the long-term stability and overall astrophysical reach of the observatories. In this work, we develop a data-driven approach that uses machine learning to model and suppress residual seismic motion within the isolation system. Ground and platform sensor data from the detectors are used to train a neural network that predicts platform motion driven by microseismic activity. When incorporated into the control scheme, the network's predictions yield up to an order-of-magnitude reduction in residual motion compared to conventional linear filtering methods, revealing that nonlinear couplings play a significant role in limiting current isolation performance. These results demonstrate that machine-learning-based control can provide a powerful new pathway for enhancing active seismic isolation, improving lock robustness, and extending the low-frequency observational capabilities of gravitational-wave detectors.

gr-qc↗

arXiv2025

Geometric and Orbital Control of Correlated States in Small Hubbard Clusters

Arrays of semiconductor quantum dots provide a powerful platform to design correlated quantum matter from the bottom up. We establish a predictive framework for engineering local electron pairing in these artificial molecules by systematically deploying three control levers: lattice geometry, orbital hybridization, and external electric fields. Using Hartree-Fock simulations on canonical 3D clusters from the tetrahedron (Z = 3) to the FCC lattice (Z = 12), at and near half-filling, we uncover three fundamental design principles. (i) Geometric Hierarchy: The resilience to Coulomb repulsion U is dictated by the coordination number Z, which controls kinetic delocalization. (ii) Orbital Hybridization: Counter-intuitively, inter-orbital hopping t_orb acts not as a simple suppressor of pairing, but as a sophisticated control knob that enhances double occupancy at moderate U by engineering the on-site energy landscape. (iii) Field Squeezing: An electric field robustly induces pairing by forcing charge localization, an effect most potent in low-connectivity clusters. These principles form a blueprint for deterministically targeting charge and spin correlations in quantum-dot-based quantum hardware.

cond-mat.str-el↗