Fraud control gets a blockchain audit trail here, with synthetic telecom/IoT requests, federated meta-learning, and QLoRA LLMs.
trending30
01 02 One Token Is Enough: Fingerprinting and Verifying Large Language Models from Single-Token Output Distributions arxiv.orgInstead of probing internals, this fingerprints LLMs from trivial one-token distributions and still tracks lineage through API chains.03 CHIP: Efficient Homomorphic Encryption-Based CNN Batch Inference Using Channel-Interleaved Packing with Small Rotation Key Set eprint.iacr.orgHE CNN inference paper: channel-interleaved packing and a small rotation-key set, trimming the usual encrypted-conv tax.04 Can federated learning keep its accuracy once you bolt on HE and DP, or does client heterogeneity still wreck it?05 Toward Stronger Code Watermarking: A Grammar-Driven Approach to Optimizing the Trade-off Between Quality and Detectability arxiv.orgGrammar-driven code watermarking, using syntax-aware masking and role biasing instead of the usual token twitching.06 Can you split a zkML inference proof into independent layer chunks and get real prover speedups? zkComposer says yes.07 MLQENABLER: Enabling Secure Machine Learning Queries over Encrypted Database in Cloud Computing arxiv.orgArXiv paper on MLQENABLER for ML queries over encrypted cloud databases, because plaintext is apparently too easy.08 A zkML explainer on proving inference ran correctly, with circuits, quantization, and the usual proof-generation pain.09 Can you fine-tune a model privately with off-the-shelf MPC/HE and still benchmark it on a real PPML task? PriFT says yes.10 Code-Level Cost Function Generation for Spatial Image Steganography Using RAG-Enhanced Large Language Models arxiv.orgLLMs are now writing steganography cost functions, with RAG and a self-evolving knowledge base in the loop.11 ArXiv paper on Ball-DP, a local-ball variant of DP with noise rules and robustness certificates, because global privacy was too blunt.12 Securing Deep Learning Hardware: A Survey of Side-Channel Vulnerabilities and Countermeasures arxiv.orgSurvey of deep-learning hardware side channels, mapping leakage to model, weights, and inputs, then the usual countermeasure laundry list.13 ePrint paper on RiskService, a modular risk scorer for TLS 1.3, PQ, or hybrid paths, because crypto agility needed another knob.14 Can distributed coded learning stay private, verifiable, and approximate when some workers go rogue? This paper tries.15 The Calibrated Deepfake Trust Score (CDTS): Competence-Coupled Trust Degradation Across Deepfake Detectors arxiv.orgPaper turns deepfake detector confidence into a trust score, then finds competence and calibration are welded together.16 Formalizes black-box proofs of ML model ownership and proves the classifier case hinges on self-correctability, naturally.17 Discard the Dross and Select the Essential: Pre-query Sample Selection for Black-box Membership Inference Attacks arxiv.orgEven membership inference gets a budget spreadsheet: it ranks samples before querying the model to save attack budget.18 LinkedIn’s PPRE wraps race/ethnicity estimation in MPC, DP, and homomorphic encryption, turning fairness checks into a privacy puzzle.19 Formal impossibility result for prompt-injection defenses in shared-embedding models, where token reuse ruins the clean split.20 Diffusion models for censoring evasion: FlowPaint edits network flows like images, with one prompt and fewer easy detections.21 Public Diffusion Models, Private Images: Key-Controlled Inversion for Conditional Reconstruction arxiv.orgPublic diffusion models, private images, and a key that decides who gets reconstruction, because inversion needed DRM.22 Privacy Vulnerabilities of Attention Layers in Tabular Foundation Models and Protection of High-Risk Queries arxiv.orgStudies membership leakage in tabular foundation models via attention, then adds an inference-time k-anonymity style defense. Lovely.23 HADES: Privacy-Preserving Federated Learning via Selective Feature Encryption and Hybrid Model Fusion arxiv.orgFederated learning, now with selective feature encryption instead of the usual full-HE tax, plus a plaintext model and fusion.24 Poisoning can make LLMs cough up training records they never saw, with a loss-landscape attack that also pokes at DP defenses.25 NeuroImprint turns federated PEFT updates into a memorization side channel, letting a malicious server recover training examples without...26 Can DP in federated learning hide backdoors? This paper says yes, with RING, and the usual defenses miss it.27 AWS shows FHE inference on SageMaker with Concrete-ML, so the cloud can now host ciphertext and call it a feature.28 Rényi-DP bounds for releasing GP posterior samples with private covariates and responses, plus membership-inference sanity checks.29 Giskard : Byzantine Robust and Confidential Aggregation for Large-Scale Decentralized Learning arxiv.orgPaper on Giskard, a tree-based MPC protocol for confidential Byzantine-robust aggregation, because committees scale so well.30 Can synthetic canaries actually strengthen privacy audits, or just give auditors prettier graphs?