arXiv subjects

Hemanth Hegadehalli Madhavarao

Publications and source records attributed to Hemanth Hegadehalli Madhavarao.

3 recordsLinked to original sources

arXiv2026

CM-DPO: Constraint-Margin Direct Preference Optimization for LLM Planning

Direct Preference Optimization (DPO) treats all constraint violations equally: a $1 budget overshoot and a $1,000 overshoot induce the same training signal. It is also susceptible to length and style bias when preference pairs come from different model families. We introduce Constraint-Margin DPO (CM-DPO), which replaces DPO's binary preference signal with a continuous margin derived from a deterministic symbolic verifier and scaled by violation severity. Hard and soft constraints are separated through a lexicographic objective, ensuring hard constraints are never traded off against preferences. To supply CM-DPO with bias-reduced training pairs, we generate preference data through procedurally generated constraint profiles (DCCG) and minimal-edit distillation from a reasoning teacher (RT-MED), within a framework we call SynPlan-R. On TravelPlanner, NaturalPlan, and out-of-distribution PlanBench, an 8B model fine-tuned with CM-DPO achieves 89.2% pass rate and 93.4% solve rate, matching multi-agent systems at 13x lower latency while outperforming GPT-4o on unseen Blocksworld by 9.2 points.

cs.AI↗

arXiv2026

Context Contamination in LLM Analysis of Network Security Logs: Poison with Passive Prompt Injection and Mitigation Evaluation

Large Language Models are increasingly deployed in Security Operations Centers for log analysis tasks including summarization, alert triage, and threat investigation. These systems ingest logs from external-facing services and process network logs as natural language contexts to generate security insights. We demonstrate that this architectural pattern introduces a critical vulnerability: adversaries can embed prompt injection payloads in log-generating fields that persist in storage and are executed when analysts query the LLM, achieving what we term passive prompt injection. We present LogInject, a systematic framework for evaluating these threats. Using LogInject-1.0, a benchmark of 12,847 log entries including 2,569 adversarial samples, we evaluate three production LLMs across four attack objectives: activity concealment, false positive generation, information exfiltration, and output hijacking. Our findings reveal an up to 88.2% attack success rate (83.4% average across models) under the baseline conditions. We introduce Context Stitching, a novel technique that fragments payloads across multiple log entries to evade stateless filters while exploiting LLM long-context reasoning, achieving a 76.4% success rate. As mitigation, we evaluate layered defenses by combining input filtering, prompt hardening, and output validation, demonstrating a 90.4% attack reduction, although 8.4% residual vulnerability persists. Our results establish that LLM-based log analysis creates an inherent confused deputy vulnerability where untrusted data and trusted instructions compete indistinguishably for model attention, requiring defense in-depth architectures and continued human oversight for security-critical decisions.

cs.CR↗

arXiv2025

QuCoWE Quantum Contrastive Word Embeddings with Variational Circuits for NearTerm Quantum Devices

We present QuCoWE a framework that learns quantumnative word embeddings by training shallow hardwareefficient parameterized quantum circuits PQCs with a contrastive skipgram objective Words are encoded by datareuploading circuits with controlled ring entanglement similarity is computed via quantum state fidelity and passed through a logitfidelity head that aligns scores with the shiftedPMI scale of SGNSNoiseContrastive Estimation To maintain trainability we introduce an entanglementbudget regularizer based on singlequbit purity that mitigates barren plateaus On Text8 and WikiText2 QuCoWE attains competitive intrinsic WordSim353 SimLex999 and extrinsic SST2 TREC6 performance versus 50100d classical baselines while using fewer learned parameters per token All experiments are run in classical simulation we analyze depolarizingreadout noise and include errormitigation hooks zeronoise extrapolation randomized compiling to facilitate hardware deployment

quant-ph↗