Open Access. Powered by Scholars. Published by Universities.®
Articles 1351 - 1380 of 4203
Full-Text Articles in Entire DC Network
Cryptographic Attestation Of Tool Responses For Gating Autonomous Agent Actions, Anonymous
Cryptographic Attestation Of Tool Responses For Gating Autonomous Agent Actions, Anonymous
Defensive Publications Series
Tool services generate cryptographically signed tool-call receipts that bind a tool-call request and a tool-call response to metadata including a caller identifier, timestamp, and nonce. The receipt includes a request hash and a response hash, and is signed (e.g., Ed25519) over receipt fields. An autonomous agent forwards the signed receipt with a subsequent action request to a downstream gate service. The gate service verifies the signature and enforces gating conditions including request binding to the targeted resource, freshness via the timestamp, and non-replay via nonce consumption. Consequential actions are authorized only when required receipt prerequisites validate; missing receipts, mismatched request …
Distributed Ledger-Based Provenance Tracking And Verification For Multi-Agent Ai Knowledge Systems, Anonymous
Distributed Ledger-Based Provenance Tracking And Verification For Multi-Agent Ai Knowledge Systems, Anonymous
Defensive Publications Series
A blockchain-based provenance verification layer is described for text-based knowledge artifacts shared among multiple AI agents. Middleware intercepts creation and consumption of artifacts in shared knowledge stores and associates each artifact or claim with a provenance record on a distributed ledger, including content hash, author identifier, timestamp, verification status, parent links, propagation depth, and confidence. Independent verification agents generate cryptographically signed attestations using techniques such as command help checks, API test calls, cross-references, and consistency checks, and ledger status is updated accordingly, optionally blocking unverified writes. During reads, a trust score is computed from verification status, attestation density, propagation depth, …
Agent Preference Memory Integrity Verification System For Detecting And Recovering From Preference Poisoning Attacks, Anonymous
Defensive Publications Series
Systems and methods are described for integrity verification of AI agent preference memory. Preference entries include semantic content, confidence, timestamp, source interaction identifier, and reinforcement history, and are associated with cryptographic provenance signatures and interaction context hashes. A preference consistency graph computes embedding-based consistency weights between preferences and produces anomaly scores for candidate preferences based on contradictions with stored high-confidence preferences. Confidence values may decay over time and be re-verified using subsequent behavior and a multi-source corroboration ladder. The system creates cryptographically signed checkpoints and performs targeted rollback to surgically remove unverifiable or anomalous preference entries while preserving verified entries, …
Recommendation Circuit Breaker System For Agent-Speed Feedback Loop Stability, Anonymous
Recommendation Circuit Breaker System For Agent-Speed Feedback Loop Stability, Anonymous
Defensive Publications Series
Systems and methods are disclosed for stabilizing recommendation rankings subject to fast feedback-loop dynamics. A circuit breaker layer monitors ranking position time series to compute ranking velocity and a catalog-level ranking volatility index, detects oscillations via frequency-domain analysis, and measures exposure concentration using concentration indices. The system may additionally detect absolute quality drift using calibration probe items and a probe-to-organic engagement gap. A composite instability score selects among graduated intervention tiers, including diversity-weight adjustment, ranking velocity capping, ranking freezes using cached stable rankings, and fallback rankings based on longer-horizon popularity. The system may classify sessions as agent-originated or human-originated using …
Position-Dependent Psychological Risk Management For Recommendation Ranking, Anonymous
Position-Dependent Psychological Risk Management For Recommendation Ranking, Anonymous
Defensive Publications Series
A recommendation system constructs a ranked slate using position-dependent psychological risk management. The system partitions slate positions into contiguous zones including a Trust zone, an Engagement zone, and a Discovery zone, with zone boundaries adapted per user based on a user trust score and optionally cognitive mode or session progress. A base ranking model produces base value scores and an uncertainty measure. For each target position, the system computes a position-modified score by applying a zone-specific risk tolerance function conditioned on user state. In the Trust zone, items are penalized as a function of prediction uncertainty scaled by user distrust; …
Feedback Loop Intensity Monitoring And Automatic Intervention For Recommendation Systems, Anonymous
Feedback Loop Intensity Monitoring And Automatic Intervention For Recommendation Systems, Anonymous
Defensive Publications Series
Systems and methods are described for monitoring recommendation feedback loops and automatically intervening in live ranking. A feedback loop intensity (FLI) for a user and topic is computed as a ratio of a user-conditional topic serving proportion to a global topic serving proportion over a rolling window. An organic interest score is computed from non-recommendation-channel evidence including one or more of search, direct navigation, creator following, or external engagement. An adjusted FLI is generated by dampening the raw FLI as a function of organic interest. When adjusted FLI exceeds configurable thresholds, a controller applies topic-specific, tiered interventions in the ranking …
Unified Dynamic User State Vector With Cross-State Conflict Resolution For Recommendation Ranking, Anonymous
Unified Dynamic User State Vector With Cross-State Conflict Resolution For Recommendation Ranking, Anonymous
Defensive Publications Series
Systems and methods are disclosed for recommendation ranking using a unified, explicitly decomposed user psychological state vector. A coordination layer loads per-user state values including mood, trust, cognitive mode, topic-indexed satiation, and topic-indexed curiosity, constructs explicit cross-state interaction features (including trust×curiosity, mood×satiation, mode×satiation, and trust×mood), and provides these to a ranking model to compute base scores for candidate items. A conflict resolution engine detects contradictory ranking actions implied by different state dimensions and applies a dominance hierarchy with cascading hard and soft constraints, including mood-based content suppression and trust-gated exploration as hard constraints and satiation discounting as a soft constraint. …
Coordination-Resistant Differential Privacy Layer For Recommendation Platform Signals, Anonymous
Coordination-Resistant Differential Privacy Layer For Recommendation Platform Signals, Anonymous
Defensive Publications Series
A recommendation platform serves signals such as rankings, trending indicators, popularity metrics, price distributions, and market-level statistics to autonomous agents. A differential privacy layer generates agent-specific responses by applying calibrated (e, d)-differential privacy noise independently per agent, optionally with per-agent consistency within an epoch, so different agents querying the same signal receive different values. The layer maintains per-agent privacy budgets with signal-type-specific query costs and composition-aware accounting, and can rate-limit or return maximally noised coordination-relevant signals upon budget exhaustion. The layer may also randomize signal freshness by scheduling per-agent update delays within signal-type-specific windows. A coordination-risk monitor computes a segment …
Negotiation-Mediated Recommendation Platform For Autonomous Agent Commerce, Anonymous
Negotiation-Mediated Recommendation Platform For Autonomous Agent Commerce, Anonymous
Defensive Publications Series
A platform mediates commerce between autonomous buyer and seller agents by treating recommendation as structured negotiation. The platform receives structured buyer requirements including hard constraints and preference information, and structured seller offers including multi-attribute flexibility ranges for terms such as price, delivery, warranty, and bundling. Compatibility scores are computed to select feasible matches and initiate a multi-round proposal and counter-proposal protocol. When negotiations stall, the platform computes or approximates Pareto-efficient candidate agreements and may select a bargaining-based compromise point. The platform may apply surplus-based fees and incentive-compatible transfer logic and may maintain credibility scores from observed negotiation behavior. Negotiation outcomes …
Evaluator Stress Testing For Recommendation Quality Gaming Detection, Anonymous
Evaluator Stress Testing For Recommendation Quality Gaming Detection, Anonymous
Defensive Publications Series
Systems and methods are described for stress-testing automated quality evaluators used in recommendation ranking to detect format-driven score gaming. For an item, the system generates format-perturbed variants that preserve semantic content and content-perturbed variants that preserve formatting, validates the variants using similarity and entailment or divergence constraints, and obtains evaluator scores for the item and variants. Format sensitivity and content sensitivity are computed from score differences, and a gaming score is derived as a ratio of sensitivities with an epsilon term. The gaming score may be used to compute an adjusted ranking score using category-calibrated parameters, to monitor item-, producer-, …
Creator Incentive Preservation System Under Agent-Mediated Consumption In Recommendation Platforms, Anonymous
Creator Incentive Preservation System Under Agent-Mediated Consumption In Recommendation Platforms, Anonymous
Defensive Publications Series
Systems and methods are described for operating a recommendation platform with creator-incentive preservation under agent-mediated consumption. The platform computes a per-creator Creator Activity Index combining production trends, quality trends, response sensitivity indicative of gaming, return-on-effort trajectory, and churn risk estimated via survival modeling. A platform Ecosystem Diversity Score is computed from active creator count, traffic concentration, topic entropy, and new-creator entry success. An inverse-relationship monitor estimates an effective recommendation-accuracy parameter from traffic distributions and identifies operation relative to a threshold regime in which increasing accuracy corresponds to decreasing creator production volume. A policy controller updates ranking using constrained optimization, including …
Performative Stability Controller For Agent-Facing Recommendation Systems, Anonymous
Performative Stability Controller For Agent-Facing Recommendation Systems, Anonymous
Defensive Publications Series
Systems and methods are disclosed for stabilizing agent-facing recommendation policies subject to performative feedback. Interaction data is collected from an exposed population receiving a deployed recommendation policy and a holdout population receiving baseline recommendations. A first predictor trained on exposed data and a second predictor trained on holdout data are compared to compute a performative effect magnitude, including a divergence between predicted distributions. Counterfactual demand estimation produces item-level recommendation dependence scores and contamination ratios separating organic and induced demand. A self-referential loop detector evaluates directionality using correlation and Granger causality and may apply Fourier analysis to detect feedback resonance. Stability …
Parasocial Relationship Lifecycle Modeling For Creator Recommendations With Phase-Specific Ranking Strategies, Anonymous
Defensive Publications Series
Techniques are described for recommending creator content using parasocial relationship lifecycle modeling. Interaction data are used to compute signals for each user-creator pair including cross-topic engagement consistency, engagement differential versus other creators within the same topics, and catalog coverage. A pair may be classified as parasocial based on thresholding these signals. A state machine assigns a lifecycle phase selected from discovery, rising attachment, peak attachment, fatigue, recovery, and disengagement using engagement level and engagement trends over time. Per-request ranking applies phase-specific strategies and exposure limits, selects a diversity dimension between topic diversity and creator diversity based on whether the user …
Reactance-Aware Adaptive Personalization Intensity With Behavioral Detection And Dynamic Modulation In Recommendation Systems, Anonymous
Defensive Publications Series
A recommendation system detects psychological reactance using explicit, implicit, and contextual behavioral signals and computes a per-user reactance score with temporal decay. A personalization intensity is determined per user and per content domain based on a domain-specific base intensity, the reactance score, and a context multiplier, subject to a minimum floor. The system modulates ranking in accordance with the intensity by adjusting feature-group weights and/or masking tiers of features. The system also generates an explanation whose framing corresponds to the computed intensity and truthfully reflects factors used for selection, including personal behavioral framing at higher intensities, neutral interest-based framing at …
Challenge Preference Generation And Boundary Testing For Filter Bubble Prevention In Content Recommendation Systems, Anonymous
Defensive Publications Series
Systems and methods are disclosed for managing negative preference boundaries in content recommendation. A boundary record is identified from explicit signals (e.g., natural-language dislikes, hides) or implicit signals (e.g., repeated exposure with low engagement) and stored with a confidence score and timestamps. Boundary confidence decays over time using an exponential decay with a floor. When a boundary’s decayed confidence meets a testing criterion, the system generates semantically adjacent challenge preferences using an ontology, including intersection bridges with liked topics, softened variants, or meta-content framings, and selects bridges using a plausibility score. One or more probe items matching a selected challenge …
Multi-Modal Ai Platform For Integrated Manufacturing Build Monitoring And Automated Defect Analysis, Anonymous
Multi-Modal Ai Platform For Integrated Manufacturing Build Monitoring And Automated Defect Analysis, Anonymous
Defensive Publications Series
A multi-modal manufacturing build monitoring platform ingests time-stamped functional test station results and inspection imagery from multiple factory stations. The platform stores the data in a unified store linked by unit identifiers and provides build-level, station-level, and device-level monitoring views. Statistical drift detection flags shifts in test result distributions, including capability metric changes and distribution shape changes. Computer vision models analyze AOI/SMT and other inspection images to detect visual defects such as solder anomalies, component placement issues, and chip/package damage. A correlation engine links visual defect patterns with numerical test trends across stations and units and generates ranked failure modes …
Consumer-Oriented Marketplace Intelligence And Auction Optimization Systems For Digital Advertising Platforms, Anonymous
Defensive Publications Series
Systems and methods are disclosed for consumer-centric marketplace intelligence in a digital advertising platform. Event data and social graph data are used to maintain a continuously updated consumer state vector per user including RFM scoring, lifecycle stage, predicted 13-month lifetime value tier, purchase propensity, and shopping rhythm. A per-user, per-category lifecycle model maintains probabilistic stage membership across stages including Unaware, Discovery, Research, Intent, Active, Champion, and Lapsed, and may represent cross-category migration sequences. A global product catalog knowledge graph normalizes products across multiple advertisers and stores cross-merchant relationships including complements, substitutes, and upgrade paths. An auction service computes auction scores …
Sampling-Based Ground Truth Discovery For Data Asset Compliance Verification, Anonymous
Sampling-Based Ground Truth Discovery For Data Asset Compliance Verification, Anonymous
Defensive Publications Series
Sampling-based ground truth discovery for data asset compliance verification is disclosed. Regulated user identifiers are sampled from one or more monitored populations, including production sampling and/or controlled test users. For multiple target data assets across heterogeneous data store types, scannable columns are identified using heuristics, machine-learning predictions, and/or semantic annotations. Join queries are executed between sampled identifiers and target assets, including nested structures such as maps, arrays, and JSON, to obtain empirical observations of user data presence. Observations are evaluated against privacy expectations for asset groups using states including ALLOW, DISALLOW, ALLOW_EXCLUSIVE, and EXIST_DEFINITE, optionally incorporating dynamic carve-out and disallow …
Privacy-Preserving Structured Logging Architecture For Semantic Artifacts In Machine Learning Inference Systems, Anonymous
Defensive Publications Series
Structured logging techniques are described for natural-language reasoning artifacts produced during inference in LLM-based recommendation systems. A reasoning chain may be decomposed into structured fields at multiple sensitivity levels, including non-sensitive numerical fields, low-sensitivity categorical fields, a scrubbed rationale, and a raw rationale. Sensitivity classification may include a first layer detecting explicit personal identifiers using pattern matching and a second layer detecting semantically sensitive categories using a sensitive-category dictionary. Scrubbing replaces sensitive sentences with category-level abstractions. The structured fields are routed to different storage tiers with differential access controls, including audit-logged access for restricted tiers. A retention policy engine enforces …
Semantic Profile Version-Based Cache Invalidation For Machine Learning Inference Systems, Anonymous
Semantic Profile Version-Based Cache Invalidation For Machine Learning Inference Systems, Anonymous
Defensive Publications Series
Systems and methods are described for caching large language model (LLM) reasoning outputs in recommendation services using semantic profile versioning. Cache entries map a key including user identifier, user profile version, and candidate identifier to a reasoning output that includes a score, tier, and rationale. A profile version increments in one embodiment when a profile update delta is classified as a semantic change, including preference additions/removals or profile regeneration, and optionally confidence updates that cross configured thresholds, while session and timestamp updates are treated as non-semantic. A two-level cache may be used with a session cache and a cross-session cache. …
Cross-Platform Agent Contamination Detection And Containment, Anonymous
Cross-Platform Agent Contamination Detection And Containment, Anonymous
Defensive Publications Series
Cross-platform agent contamination detection and containment is disclosed for autonomous agents that operate across multiple recommendation platforms while sharing preference memory. Preferences are stored with structured provenance metadata identifying an origin platform and interaction context, optionally including cryptographic signatures, confidence tiers, influence tracking, and quarantine status. Cross-platform monitoring computes temporal and categorical correlation metrics, including influence scoring and causality-oriented tests, to determine whether acquisition on one platform predicts behavior change on another. When suspected contamination is detected, affected preferences are quarantined with platform-differentiated influence control, optionally including recursive quarantine of influenced preferences, while enabling independent local verification and cross-verification to …
Dynamic Multi-Space Agent Rotation For Persistent Ai Assistants In Rotating Operational Roles, Anonymous
Dynamic Multi-Space Agent Rotation For Persistent Ai Assistants In Rotating Operational Roles, Anonymous
Defensive Publications Series
A single persistent AI agent operates across multiple operator-dedicated communication spaces and switches which space it actively serves according to external rotation schedule data. The agent maintains a unified domain knowledge base that persists across rotations and separate per-user profiles that customize behavior per operator. At rotation boundaries, the agent automatically generates a shift summary for an outgoing operator, stores open issue context into the unified domain knowledge base, switches active output routing to an incoming operator’s space, loads the incoming operator’s profile, sends a welcome briefing including open issues and recent patterns, and activates monitoring in the new space. …
Convergent Memory Architecture - Unified Tiered Store For Embedding Vectors And Natural Language User Profiles, Anonymous
Defensive Publications Series
A unified tiered store serves both fixed-size embedding vectors and variable-size natural-language user profiles from a shared hierarchy including GPU HBM, host DRAM, and SSD-backed persistence. A single database instance maintains separate logical partitions for embeddings and profiles with different table formats and caching representations. Cache capacity is dynamically partitioned between data types based on observed workload rates and expected value sizes with smoothing and minimum reservations. A heterogeneous I/O scheduler prioritizes embedding SSD reads and reserves SSD service for embeddings while remaining work-conserving for profile reads. Cross-type cache coherence is provided by tracking associations between user identifiers and derived …
Write-Amplification-To-Cache-Warming Conversion For Collaborative Memory Propagation In Recommendation Systems, Anonymous
Defensive Publications Series
Systems and methods are described for converting collaborative memory propagation write activity into selective cache warming for recommendation-serving data. Upon a propagation event that writes preference information for a target user, a controller estimates a warming value based on a return probability for the target user, cache hit probability, storage latency, and request frequency. The return probability may be elevated using social clustering signals including co-activity, tie strength, and temporal patterns. Subject to a storage bandwidth warming budget (e.g., token-bucket limited and traffic-health adaptive), the controller issues an asynchronous warming read that loads the target user embedding or profile data …
Workload-Parameterized Performance Prediction Framework For Embedding Cache Optimization Triage, Anonymous
Workload-Parameterized Performance Prediction Framework For Embedding Cache Optimization Triage, Anonymous
Defensive Publications Series
A workload-parameterized performance prediction framework is described for triaging embedding cache optimizations across a fleet of machine-learning models. Workload parameters including a distribution parameter (alpha), reuse, table size N, batch size, and embedding dimension are input to a closed-form causal-chain model that predicts unique indices, cache misses across multiple cache levels, persistent-store reads, and storage I/O operations. A predicted latency is computed as a weighted combination of predicted access counts and latency terms, and a bottleneck class is identified. Each optimization is associated with a targeted segment of the causal chain and a maximum efficiency, enabling estimation of predicted speedups …
Workload-Parameterized Performance Prediction Framework For Embedding Cache Optimization Triage, Anonymous
Workload-Parameterized Performance Prediction Framework For Embedding Cache Optimization Triage, Anonymous
Defensive Publications Series
A workload-parameterized performance prediction framework is described for triaging embedding cache optimizations across a fleet of machine-learning models. Workload parameters including a distribution parameter (alpha), reuse, table size N, batch size, and embedding dimension are input to a closed-form causal-chain model that predicts unique indices, cache misses across multiple cache levels, persistent-store reads, and storage I/O operations. A predicted latency is computed as a weighted combination of predicted access counts and latency terms, and a bottleneck class is identified. Each optimization is associated with a targeted segment of the causal chain and a maximum efficiency, enabling estimation of predicted speedups …
Zipf-Aware Adaptive Cache Hierarchy Configuration For Storage-Backed Embedding Systems, Anonymous
Zipf-Aware Adaptive Cache Hierarchy Configuration For Storage-Backed Embedding Systems, Anonymous
Defensive Publications Series
Techniques are described for configuring a multi-tier cache hierarchy for SSD-backed embedding stores based on measured or predicted Zipf access skew. Live embedding index accesses are monitored to estimate a Zipf exponent using a sliding window and a Hill estimator, and temporal reuse may be estimated from inter-batch overlap. Cache sizes for multiple tiers are derived using closed-form Zipf working-set relationships for target hit rates, including handling for exponents greater than one and approximations near one. Cache sizes may be dynamically reconfigured at runtime using smoothed exponent estimates and a deviation threshold, without restarting processing; evicted entries from an upper …
Multi-Organ Living Donor Matching System Using Social Graph Embedding With Donation-Aware Bias And Heuristic Medical Compatibility Scoring, Anonymous
Defensive Publications Series
A computer-implemented platform matches living organ donors with recipients using self-reported profile data and social-network data. A request is routed by organ type to kidney, liver, or bone marrow logic. A Friend2Vec graph embedding model constructs a weighted social graph, performs biased random walks, and trains a Skip-Gram network to produce user embeddings. Friend2Vec adds a donation-aware bias parameter that preferentially traverses edges associated with successful donation events. Embeddings are fused with medical and geographic attribute vectors using a learned fusion layer with layer normalization. A multi-layer perceptron predicts P(Donation | Donor, Recipient) from concatenated donor and recipient embeddings, their …
Dual-Mode Adaptive Cache Management For Storage-Backed Embedding Systems, Anonymous
Dual-Mode Adaptive Cache Management For Storage-Backed Embedding Systems, Anonymous
Defensive Publications Series
Techniques are described for adaptive cache management in SSD-backed embedding table systems that serve both training and inference. A mode detection engine monitors access pattern features including access multiplicity, write frequency, batch sequentiality, and request concurrency, and classifies operation as training, inference, or hybrid. Based on the classification, a cache manager switches among mode-specific policies for locking, prefetch, eviction, and write-back. Training mode may use generation-based locking, DataLoader lookahead prefetch, distance-based eviction, and write-back of dirty entries on eviction. Inference mode may use reference-count locking to block eviction during active requests, request coalescing with index deduplication for prefetch, and frequency-weighted …
Gpu-Storage I/O Isolation For Distributed Embedding Training Systems, Anonymous
Gpu-Storage I/O Isolation For Distributed Embedding Training Systems, Anonymous
Defensive Publications Series
A GPU-storage I/O isolation architecture is described for distributed embedding training with SSD-backed embedding tables. Storage-related callbacks are registered using a host function launch mechanism that holds a GPU driver mutex only during enqueue, allowing blocking key-value store reads to execute on a CPU thread without stalling other CUDA streams. SSD read completions are delivered through a condition-variable-backed fill queue that wakes a filler thread without polling. Cache eviction is overlapped with prefetch using a double-buffer eviction manager that alternates buffers across training steps while a background thread writes dirty entries back to SSD. A stream isolation scheduler assigns dedicated …