Open Access. Powered by Scholars. Published by Universities.®
- Publication Year
- File Type
Articles 1141 - 1170 of 11790
Full-Text Articles in Entire DC Network
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. …
Runtime-Constructed Execution Graphs With Agent-Directed Path Selection And Constraint Enforcement For Serving Systems, Anonymous
Defensive Publications Series
A serving system constructs and executes a directed acyclic graph (DAG) at runtime for each request based on an agent model’s tool-call decisions. The DAG is incrementally formed from base tools and intent-selected tools chosen using request intent, conversational context, and intermediate results. Runtime safety constraints are enforced, including a maximum path length, per-tool repetition bounds to avoid cycles, allowlist enforcement, and a token budget. A tool registry provides per-tool description, capability tags, and estimated token and latency metadata that may be used to filter or prioritize tool selection under remaining budgets. Execution may include multiple phases, with later phases …
Write Amplification Management For Machine Learning-Based Collaborative Propagation Systems, Anonymous
Write Amplification Management For Machine Learning-Based Collaborative Propagation Systems, Anonymous
Defensive Publications Series
Techniques are described for managing write amplification in recommendation systems that use large language models (LLMs) to semantically extract preference insights from user interactions and to propagate insights to similar users. A propagation controller applies hierarchical quotas including interaction-level confidence gating, per-user limits, and a system-wide budget. Neighbor selection uses graph-topology-aware fan-out reduction based on user out-degree in a similarity graph, with neighbors ranked by similarity and top-K selected. Work is scheduled into a two-tier priority queue in which fresh interaction processing is handled with strict priority over propagated updates. Low-priority propagation tasks may be temporally batched for coalescing. A …
Hybrid Serving Architecture With Cost-Benefit Routing For Llm-Based Recommendation Systems, Anonymous
Hybrid Serving Architecture With Cost-Benefit Routing For Llm-Based Recommendation Systems, Anonymous
Defensive Publications Series
A hybrid online serving architecture for recommendation systems selectively applies large language model (LLM) reasoning at a late ranking stage. A request is processed through retrieval and early-stage ranking to reduce candidates and produce confidence statistics. A cost-benefit routing orchestrator computes a routing score using traditional-model confidence, user value tier, budget utilization, latency constraints, and surface priority, and compares the routing score to a dynamically adapted threshold responsive to budget pressure, quality gap, and tail-latency pressure. Prior to LLM invocation, an ROI estimate is computed from expected quality lift and token cost, and the LLM is invoked only when ROI …
Shadow Scoring System For Cost-Effective Llm Recommendation Experimentation, Anonymous
Shadow Scoring System For Cost-Effective Llm Recommendation Experimentation, Anonymous
Defensive Publications Series
A shadow scoring framework evaluates LLM-based recommendation ranking without serving LLM-ranked results to users. For enrolled requests, a production ranker serves a primary ranking while an LLM ranker computes a shadow ranking asynchronously off the serving critical path. The system logs paired observations including the primary ranking, the shadow ranking, and observed engagement with the served slate. Multi-level enrollment controls cost using persistent user hashing at a configurable rate, per-user request subsampling, and a daily token budget cap that pauses shadow scoring when exceeded. Counterfactual metrics are computed from logs, including NDCG comparison using observed engagement as relevance, promoted item …
Per-User Kv Cache With Active User Windowing And Tiered Memory Placement For Billion-Scale Llm Recommendation, Anonymous
Defensive Publications Series
Techniques manage per-user attention key-value (KV) caches for large language model (LLM) recommendation serving at very large user scale. A cache manager maintains activity windows to bound the number of users whose KV cache tensors are retained. KV caches for users active within a hot window are stored in GPU high-bandwidth memory at higher precision, while KV caches for users active within a warm window are stored in host RAM in a more compact quantized format. Entries are demoted from the hot tier to the warm tier after a hot inactivity threshold using quantization, promoted back to the hot tier …
Token-Aware Capacity Management With Multi-Level Budget Hierarchy For Llm-Based Recommendation Systems, Anonymous
Defensive Publications Series
Token-aware capacity management is described for LLM-based recommendation serving. A request token cost is estimated as a sum of token components including profile, candidate, context, and output tokens. Admission control evaluates the estimated cost against a four-level hierarchical token budget comprising a global budget, per-surface budgets, per-user budgets derived from user tiers, and a per-request cap. For sub-100 ms serving paths, serving instances perform budget checks and deductions using local leased counters obtained asynchronously from a central token pool, enabling constant-time decisions without network round-trips and non-blocking lease renewal. When budgets are insufficient, a graceful degradation orchestrator selects among full …
Computer Vision Camera Sensor With Embedded Ai Features And Imu For Ultra-Low Power Adaptive Device Management, Anonymous
Defensive Publications Series
A computer vision camera sensor module includes an image sensing element, embedded AI processing, and an inertial measurement unit (IMU) to support ultra-low power always-on operation and adaptive device management. The sensor executes AI algorithms locally to analyze visual data and IMU motion data, detect user activity and environmental change, and select among tracking modes including 3-DoF tracking for sedentary activity and 6-DoF tracking for standing or walking. Based on detected meaningful change, the sensor autonomously maintains other device sensors or cameras in standby or triggers their activation, while avoiding wake-up of a backend processor for decision-making. The module may …
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 …
Modular Protocol Bridge Band For Enabling Certified Smartphone Communication In Neural Interface Wearables, Anonymous
Defensive Publications Series
A replaceable wearable band includes an embedded protocol bridge module that enables certified smartphone communication for a wearable core device. The band carries a wired interface that provides power and data between the wearable core device and the protocol bridge module. The protocol bridge module includes an authentication coprocessor and a dual-mode Bluetooth system-on-chip and translates between a wired-device protocol and a certified accessory communication protocol carried over Bluetooth Classic to a smartphone. The wearable core device interacts with the protocol bridge module as a wired peripheral without implementing the certified accessory protocol. In some embodiments, the protocol bridge module …
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 …
Dual-Mode Cache Observability With Feedback-Driven Optimization For Machine Learning Embedding Stores, Anonymous
Defensive Publications Series
Techniques are described for cache observability in machine-learning embedding stores using dual-mode measurement with non-interfering consumers. A monotonic, non-resetting vector of counters tracks total lookups and tier-specific hits. Monitoring derives cumulative hit rates via non-mutating reads of the counters. Profiling derives windowed hit rates by capturing counter snapshots at boundaries and computing deltas, enabling virtual resets and overlapping windows without resetting shared state. The same counters are exposed across multiple software layers (e.g., native code, graph execution, and scripting) using scalar loads and simple arithmetic. A feedback controller maintains a ring buffer of per-batch deltas to compute sliding-window hit rates, …
Pipelined Asynchronous Prefetch With Decoupled Fill Queues And Completion-Order Processing For Storage-Backed Embedding Tables, Anonymous
Defensive Publications Series
Techniques are described for prefetching embeddings from SSD-backed storage using a decoupled pipeline. Cache misses are identified by a caller thread and enqueued to a miss queue. A dedicated reader stage dequeues misses, issues asynchronous multi-key storage reads, and forwards completed results individually to a fill queue in completion order rather than request order. A dedicated cache filler stage dequeues fill entries and inserts embedding values into one or more caches, optionally selecting fill work using a priority policy that accounts for urgency (e.g., embeddings awaited by a forward pass), access frequency, and batch position. The reader stage may adapt …
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 …