Open Access. Powered by Scholars. Published by Universities.®
- Publication Year
- File Type
Articles 1231 - 1260 of 11791
Full-Text Articles in Entire DC Network
Dynamic K-Anonymity, Anonymous
Dynamic K-Anonymity, Anonymous
Defensive Publications Series
Techniques are disclosed for privacy enforcement in streaming data aggregation. Contextual signals for an aggregation breakdown are obtained, including signals such as breakdown cardinality, traffic volume, targeting specificity, data age, regulatory tier, and cross-breakdown overlap. A risk score in a bounded range is computed and mapped to a dynamic k-anonymity threshold between configured minimum and maximum values, optionally adjusted for composition risk based on how many breakdowns a user contributes to. An existing k-anonymity check uses the dynamic threshold to gate release. Separately or in a closed loop, probabilistic cardinality sketches are monitored over time to forecast whether cohorts will …
Hypernetwork Cross-Attention, Anonymous
Hypernetwork Cross-Attention, Anonymous
Defensive Publications Series
Techniques are described for conditioning beam-search decoding on encoder outputs without constructing cross-attention matrices over encoder positions. An encoder output is compressed once into a fixed-size summary vector using learned query vectors. For each beam, a decoder query is mapped to a low-dimensional latent code by a code generator. A hypernetwork generates parameters for a per-beam primary network as a function of the latent code, including affine weight generation. The primary network is applied to the shared summary to produce a beam-specific output vector, with per-beam computation independent of encoder sequence length. A continuous latent code space supports interpolation, extrapolation, …
Pareto Beam Search, Anonymous
Pareto Beam Search, Anonymous
Defensive Publications Series
Techniques are described for multi-objective beam search in generative retrieval systems that autoregressively generate semantic identifiers across multiple codebook steps. Each beam carries a vector of objective scores, such as relevance, diversity, freshness, and business value. At each decoding step, candidate beam extensions are evaluated and pruned by Pareto dominance using non-dominated sorting to form fronts. Beam selection proceeds by taking fronts in order until a beam budget is reached, and crowding distance is used to select among candidates within a front to promote spread across objective space. In some implementations, each beam also carries a preference vector used to …
Composed Ot Beam Search, Anonymous
Composed Ot Beam Search, Anonymous
Defensive Publications Series
Multi-codebook generative retrieval produces semantic identifiers by autoregressively generating K codebook tokens. The disclosure describes globally coordinated decoding by formulating the K-step process as a sequentially composed optimal transport problem with K transport plans linked by marginal consistency constraints. A retrieval transport cost is defined per step from decoder conditional probabilities and may include a learned future value term and a diversity penalty. An entropically regularized objective is solved using composed Sinkhorn iterations that perform forward and backward scaling passes across steps to satisfy source, intermediate, and target marginals. A global beam computation budget is adaptively allocated across steps based …
Automatic Discovery Of User Intent Categories In Conversion Rate Prediction Using Differentiable Discrete Routing, Anonymous
Defensive Publications Series
Systems and methods are described for automatically discovering user intent categories in conversion rate prediction using differentiable discrete routing. A routing network produces logits over K expert prediction heads based on interaction features such as click type, page type, and position. During training, a Gumbel-Softmax reparameterization generates differentiable approximations of discrete expert assignments, optionally with temperature annealing to transition from soft sharing to near-discrete specialization. A load balancing loss, such as a KL-divergence between an average routing distribution and a uniform distribution, discourages expert collapse. During inference, routing may collapse to a hard argmax such that exactly one expert head …
Sram-Packed Multi-Leaky-Bucket Rate Limiter With Timer-Tick Scrubbing, Pipelined Read-Modify-Write Forwarding, And Csr Priority Escalation, Anonymous
Defensive Publications Series
A hardware leaky-bucket rate limiter stores multiple bucket entries per SRAM row and accesses the SRAM using a unified pipeline that selects per cycle exactly one of a posted traffic request, a timer-driven scrub operation, or a CSR access. Request metadata is pipelined for SRAM read latency and updates are performed using read-modify-write. A conflict detector with forwarding supplies a most-recent-value view when a row is accessed again before a prior update is observable, avoiding lost updates. Scrub is driven by a programmable timer tick and supports per-bucket refill rates using a scrub frequency counter; an error indication asserts when …
Recoverable Multimedia Export With Pre-Export Validation And Element-Level Automated Remediation, Anonymous
Recoverable Multimedia Export With Pre-Export Validation And Element-Level Automated Remediation, Anonymous
Defensive Publications Series
Techniques are disclosed for recoverable multimedia export using pre-export validation and element-level automated remediation. A validation engine traverses a composition graph before an export pipeline begins and checks referenced elements such as media assets, audio tracks, fonts, effects, filters, LUTs, and external resources for existence, decodability or integrity, permissions, compatibility, dependencies, and license validity. A structured validation report identifies each problematic element with a failure type and severity and provides remediation options. A user interface presents the report prior to export, including per-issue detail and a fix-all control, and accepts user overrides. A remediation engine applies selected actions such as …
Ensemble Model Divergence Analysis For Adversarial Prompt Detection In Generative Ai Systems, Anonymous
Ensemble Model Divergence Analysis For Adversarial Prompt Detection In Generative Ai Systems, Anonymous
Defensive Publications Series
Techniques are disclosed for detecting adversarial, obfuscated, or harmful prompts in generative AI platforms by using model disagreement as a risk signal. A user prompt is routed to an ensemble of language models selected for diversity in architecture, size, generation, and safety tuning, optionally including smaller canary models. Each model generates a response, and the responses are embedded into a shared semantic vector space. Pairwise divergence values, such as cosine distances between response embeddings, are computed and aggregated into a divergence score for the prompt, with optional per-model divergence diagnostics. The divergence score is compared to a calibrated threshold to …
Simpson's Paradox Detection And Granularity-Adaptive Behavioral Curve Estimation For Recommendation Systems, Anonymous
Defensive Publications Series
Systems and methods are described for exposure-response behavioral curve estimation in recommendation systems with detection of aggregation artifacts consistent with Simpson’s paradox. Interaction logs are used to fit a behavioral curve model at multiple granularities, including individual and aggregate levels, and to compute peak locations. A distortion factor and severity score quantify divergence between aggregate and individual peak parameters and can trigger a paradox determination. A per-user peaked-behavior classifier is null-calibrated by generating synthetic sequences from a fitted monotonic model to estimate a null false-positive rate and adjust decision thresholds. A granularity-adaptive selector chooses among individual, cohort, and aggregate models …
Computational Vertical Binocular Disparity Calibration Method For Augmented Reality Displays With Perceptual Tolerance-Based Statistical Assessment, Anonymous
Defensive Publications Series
Techniques are described for calibrating vertical binocular disparity (VBD) in optical see-through augmented reality displays using two independent measurements rather than direct perceptual display-to-background alignment. A background VBD (bg_vbd) for the see-through optical path and a display-to-display VBD (display_vbd) between left and right display channels are measured, and a correction offset is computed as offset = bg_vbd - display_vbd. The offset is applied to one or both display channels to reduce relative VBD between displayed content and background toward approximately zero. A statistical framework assesses calibration quality by propagating measurement uncertainties (e.g., s_offset = sqrt(s_bg^2 + s_display^2)), optionally including display …
Active-Feedback User Simulation Agent For Bidirectional Content Retrieval In Recommendation Systems, Anonymous
Active-Feedback User Simulation Agent For Bidirectional Content Retrieval In Recommendation Systems, Anonymous
Defensive Publications Series
Techniques are described for bidirectional candidate generation in a recommendation pipeline using an active-feedback user simulation agent. An initial candidate set is retrieved using one or more retrieval paths. A user digital twin consumes a live session state, including an emotional-state embedding, and evaluates the candidates by predicting a session trajectory. When the predicted trajectory indicates low engagement, disengagement risk, or a coverage gap between predicted preference and available candidates, the agent generates a structured, machine-readable retrieval query (e.g., topic, content-type, social, negative, emotional, or mixing requests). A query handler executes the query to obtain additional candidates, forming an expanded …
Encoding-Aware Collusion Detection And Intervention System For Autonomous Pricing Markets, Anonymous
Encoding-Aware Collusion Detection And Intervention System For Autonomous Pricing Markets, Anonymous
Defensive Publications Series
Systems and methods monitor autonomous pricing markets for tacit collusion using externally observable posted prices and market outcomes. Multiple indicators are computed, including a price elevation index with Newey-West HAC uncertainty, Granger causality using vector autoregression with heteroskedasticity-consistent errors, a constrained three-state hidden Markov model producing a collusive-state posterior probability, and conditional mutual information estimated by k-nearest neighbors and normalized against a permutation null. The indicators are combined into a composite collusion score. An encoding regime is inferred from observables including price dimensionality, correlation structure, and within-run dispersion, and regime-specific thresholds are applied for detection. The system may estimate latent …
Temperature Selection For Knowledge Distillation In Mixed-Vocabulary Models, Anonymous
Temperature Selection For Knowledge Distillation In Mixed-Vocabulary Models, Anonymous
Defensive Publications Series
Plateau-aware temperature selection is described for knowledge distillation of mixed-vocabulary models that include natural-language tokens and structured identifiers (SIDs). Teacher logits for SIDs are obtained from a forward pass on calibration data and analyzed without training a student model. A cold-collapse temperature boundary is estimated as a minimum temperature at which perplexity of a temperature-scaled SID softmax exceeds a threshold, and a soft-collapse temperature boundary is estimated as a maximum temperature at which a discriminability measure based on probability ratios exceeds a threshold. A SID distillation temperature is selected from the plateau between the boundaries, including selecting a geometric mean …
Multimodal Session Simulation With Sequential User State Evolution For Content Ranking, Anonymous
Multimodal Session Simulation With Sequential User State Evolution For Content Ranking, Anonymous
Defensive Publications Series
Techniques are described for multimodal simulation of a user browsing session through an ordered content feed. A feed is rendered into a visual representation, and recent engagement events are encoded into a live session state vector that includes an emotional state embedding inferred from behavior. A multimodal autoregressive model consumes the rendered feed and the session state to predict a trajectory across feed positions, including predicted interactions and predicted emotional-state shifts conditioned on prior positions. A planner and/or safety checker uses the predicted trajectory to select among candidate orderings or to block a slate and serve a fallback ordering when …
Dual-Table Hardware Architecture For Frequency-Adaptive Memory Access Management In Accelerator Systems, Anonymous
Defensive Publications Series
A hardware trap positioned between processing elements and a memory router and/or last-level cache intercepts tagged embedding read requests. The trap includes a dual-table structure comprising a low-frequency table that tracks access frequency for addresses without storing full embedding data and a high-frequency table that stores embedding data for addresses promoted as high-frequency. When an address exceeds a configurable promotion threshold, the trap initiates a fill request and promotes the address for local servicing. During the fill window, a configurable blocking threshold stalls additional requests for the promoted address to limit backpressure, and stalled requests are serviced upon data arrival. …
Language Model System For Account Compromise Detection Via Multi-Signal Business Graph Reasoning, Anonymous
Language Model System For Account Compromise Detection Via Multi-Signal Business Graph Reasoning, Anonymous
Defensive Publications Series
Systems and methods are described for pre-monetization detection of business account compromise using large language model reasoning over business-graph evidence. A pipeline selects candidate accounts based on novel behavior such as advertising on previously unseen pages or domains. For each candidate, signals are aggregated across categories including account structural metadata, creator indicators, candidate ad content attributes, historical baseline ad attributes, administrative access modification records, and business activity events. Heterogeneous signals are transformed into a structured natural-language representation organized by category and combined with embedded domain heuristics specifying compromise indicators, non-compromise indicators, and exclusion criteria. A large language model processes the …
Neural Network-Based Feature Imputation For Privacy-Constrained Environments Using Cross-Population Training, Anonymous
Defensive Publications Series
Techniques are described for privacy-preserving feature imputation in environments where a limited-personalization policy deterministically withholds a subset of user features. Complete feature vectors from consenting users are collected and normalized using automated statistical classification of features. A deterministic binary mask derived from a privacy policy configuration is applied during training to simulate restricted-feature conditions. A denoising autoencoder (or other encoder-decoder model) is trained by reconstructing full vectors from masked inputs while computing reconstruction loss only on masked positions, optionally adding noise to visible features. During inference, visible features for a privacy-restricted user are normalized, passed through the trained model to …
Multi-Root Trust Chain Management For Secure Boot Firmware Development And Signing Infrastructure, Anonymous
Multi-Root Trust Chain Management For Secure Boot Firmware Development And Signing Infrastructure, Anonymous
Defensive Publications Series
Techniques are described for managing multiple secure-boot trust chains for firmware development and release on devices that support multiple hardware roots of trust. A provisioning pipeline generates a device configuration image that activates a selected non-production root index and is cryptographically bound to a device identifier such as an SoC serial number. The configuration image is signed using a production root to bootstrap trust, and activation may be policy-gated and generated in an event-driven manner with storage for fast retrieval. A signing service, backed by hardware security modules, signs firmware artifacts via authenticated service-to-service RPC and supports build-time selection between …
Multi-Resolution Image Segmentation With Selective Region-Based Detail Refinement, Anonymous
Multi-Resolution Image Segmentation With Selective Region-Based Detail Refinement, Anonymous
Defensive Publications Series
Techniques are disclosed for multi-stage image segmentation with selective high-resolution region refinement. A coarse segmentation neural network generates a full-frame mask while operating with a reduced internal resolution to meet real-time constraints. From coarse outputs, a region of interest (ROI), such as a head/hair region, is dynamically determined per image or per video frame. The ROI is cropped from a higher-resolution version of the original input image, optionally along with aligned coarse segmentation data, and provided to a second refinement neural network that produces refined alpha or mask values and confidence. A fusion stage combines refined ROI output with the …
Multi-Modal Social Context Detection And Automated Environment Behavior Adaptation, Anonymous
Multi-Modal Social Context Detection And Automated Environment Behavior Adaptation, Anonymous
Defensive Publications Series
A smart environment controller receives multi-modal occupancy signals including combinations of Wi-Fi sensing, Bluetooth proximity, acoustic analysis, and entry/lock events. Sensor fusion produces occupant hypotheses indicating identities, classes, and/or counts of occupants. A social context classifier determines a current social context reflecting social composition, such as solo, with-partner, hosting, kids-home, or user-defined contexts. A behavior policy engine maps the social context to coordinated actions across environment dimensions including notification routing, shared-surface content visibility, ambient settings, device routing, and security. Privacy routing rules prevent sensitive information from appearing on shared surfaces when non-primary household members are detected and instead route such …
Split-Trust Key Derivation For Passkey-Authenticated Encrypted Data Backup And Recovery, Anonymous
Split-Trust Key Derivation For Passkey-Authenticated Encrypted Data Backup And Recovery, Anonymous
Defensive Publications Series
Techniques are described for passkey-authenticated encrypted backup and recovery using split-trust key derivation. A client device performs passkey authentication via WebAuthn/FIDO2 and invokes a PRF extension to obtain a high-entropy pseudorandom output. The client derives a client-side key component (e.g., a BackupRootKey via HKDF) from the PRF output and obtains a server-side secret from a backup service only after successful passkey authentication. The client combines the client-side key component and the server-side secret using a key derivation function to produce a backup encryption key used to encrypt backup data for cloud storage and to decrypt the backup during restore. Passkey …
Quorum-Certified Partial State Recovery In Replicated Distributed Systems, Anonymous
Quorum-Certified Partial State Recovery In Replicated Distributed Systems, Anonymous
Defensive Publications Series
A replicated state machine maintains a quorum-attested committed-state shadow in addition to a consensus-replicated durable log. The shadow is organized into serviceable units (e.g., ranges, shards, or object groups) and includes, per unit, a log term/index through which the unit is safe and a deterministic digest of unit state, optionally with predecessor linkage, coverage metadata, and replica attestations. After leader failover, a replacement leader exchanges shadow records or attestations with followers and computes certified recovery frontiers per unit. In partial-recovery mode, a request gate permits reads from the shadow for certified units and may admit writes for certified units into …
Ai-Driven Contextual Terminology Definition Generation For User Interface Elements, Anonymous
Ai-Driven Contextual Terminology Definition Generation For User Interface Elements, Anonymous
Defensive Publications Series
Systems and methods generate contextual, in-product definitions for domain terminology displayed in user interfaces. A client renders an annotated term and, upon a user click, sends the term and UI context to a backend definition service. The service checks a cache keyed by the term and context; when available, a cached definition is returned. Otherwise, the service invokes a language model with references to authoritative knowledge sources and an output contract requiring a definition of 20 words or fewer and a learn hyperlink to a wiki page that defines the term. The definition and hyperlink are presented via a standardized …
Context-Guided Ai Data Activation Control System With Policy-Constrained Planning And Validation-Gated Export, Anonymous
Defensive Publications Series
A data activation control plane receives destination-side account or tenant context and a policy grant for a source environment. The policy grant is compiled into a policy-scoped planning interface that constrains tools and schema metadata available to a planning model and prevents access to raw source values. Based on the bounded destination-side context and the policy-scoped source surface, the planning model proposes a candidate activation. The control plane compiles the proposal into a structured activation plan identifying output fields, source paths, transforms, and destination parameters. A source-local validator validates the plan within the source environment, optionally including dry-run execution and …
Dual-Randomization Experimentation Framework With Routing-Based Isolation For Agent-Facing Workflow Systems, Anonymous
Defensive Publications Series
A dual-randomization experimentation framework for agent-facing workflow systems is described. Incoming work items are deterministically assigned to test or control groups at creation time, and agents are independently assigned to test or control cohorts. Isolation is enforced by integrating with a skills-based omni-channel routing engine so that test work items are routed only to test agents and control work items only to control agents. A gatekeeper controls code paths and user interface component visibility based on cohort membership, supporting phased rollouts and rollback via configuration. Group assignment is preserved across the work item lifecycle including transfers, escalations, and re-routing. Exposure …
Neuro-Symbolic Business Account Compromise Detection With Agentic Ai And Inductive Knowledge Discovery, Anonymous
Defensive Publications Series
A neuro-symbolic system detects compromise of business accounts using an agentic investigation stage and a knowledge-based decision stage. An LLM agent invokes multiple investigative tools via a model context protocol and produces an initial assessment with a rationale. The rationale and tool results are normalized into a fixed-dimensional signal vector. A hybrid FOIL and inverse entailment inductive logic programming procedure discovers interpretable confirmation and rejection rules from examples and background predicates. Rule firings and signal features are converted into multi-order features and scored using a Naïve Bayes classifier with smoothed conditional probability tables. A contradiction solver identifies disagreement patterns among …
Attractor Surgery For Post-Training Repair Of Collusive Pricing Policies Via Targeted Q-Entry Editing, Anonymous
Defensive Publications Series
Disclosed techniques perform post-training repair of reinforcement-learning pricing agents by targeted editing of action-value information to eliminate high-price attractor cycles. A frozen policy graph is constructed from a greedy policy derived from Q-values, and the graph is decomposed to identify attractor cycles. Each cycle is scored using an elevation metric relative to benchmark prices and compared to a threshold to classify high-price attractors. For states on a high-price cycle, the greedy collusive action is replaced with a one-shot best-response action by demoting Q(s,a_collusive) and promoting Q(s,a_BR) by an editing margin that flips the argmax, while leaving all other Q-entries unchanged. …
Distributed Reconfigurable Logic Cells For Zero-Respin Post-Tapeout Silicon Defect Recovery, Anonymous
Distributed Reconfigurable Logic Cells For Zero-Respin Post-Tapeout Silicon Defect Recovery, Anonymous
Defensive Publications Series
Distributed reconfigurable logic cells are integrated across a system-on-chip and/or multi-chiplet package to enable post-tapeout remediation of defects in hardened circuitry without redesign. Small micro-FPGA cells are placed proximate to selected hardened IP blocks and coupled via local interfaces and steering circuitry. After fabrication, a configuration controller programs one or more proximate cells at boot time and/or runtime with remedial logic that replaces, augments, or bypasses defective behavior, and activates routing/steering to apply the fix. Unused cells may be power-gated or clock-gated to reduce static and dynamic power. The cells may also be repurposed for debug, test, monitoring, security patches, …
Smart Composer With Graduated Ai Engagement Modes For Business Customer Conversations, Anonymous
Smart Composer With Graduated Ai Engagement Modes For Business Customer Conversations, Anonymous
Defensive Publications Series
A business messaging smart composer provides a composer-native mode switcher for selecting a thread-specific level of AI participation. A selected mode persists per conversation thread and controls AI behavior. In a manual mode, the user composes messages without AI draft insertion. In a suggested mode, an AI-generated suggested reply card presents a preview that can be tapped to load content into the composer for editing prior to user-authorized sending, with a coaching card shown when no useful suggestion is available. In an AI reply mode, AI-generated replies are automatically sent, with handoff to a human user when handling criteria are …
Unified Property Declaration System For Entity-Component State And World-Scope Governance Records, Anonymous
Unified Property Declaration System For Entity-Component State And World-Scope Governance Records, Anonymous
Defensive Publications Series
A unified property declaration system is described for collaborative environments that maintain both entity-component state and world-scope governance records. A single declaration surface defines properties for component types attachable to entities and for world-scope record kinds using a common type system and attributes. Mutations to either category are encoded as operations in a discriminated-union format whose address selects between entity-component-property addressing and world-scope record addressing. All operations traverse a unified pipeline providing consistent ordering, conflict handling, undo/redo, persistence via logging and snapshots, replication to clients, and audit logging. Compile-time scope enforcement prevents cross-registration of component types and world-scope record kinds. …