File size: 11,898 Bytes
40d7073 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 | /**
* Neural Embedding System - Frontier Embedding Intelligence
*
* Implements late-2025 research concepts treating embeddings as:
* 1. CONTROL SIGNALS - Semantic drift detection, reflex triggers
* 2. MEMORY PHYSICS - Forgetting curves, interference, consolidation
* 3. PROGRAM STATE - Agent state management via geometry
* 4. COORDINATION PRIMITIVES - Multi-agent swarm alignment
* 5. SAFETY MONITORS - Coherence detection, misalignment alerts
* 6. NEURAL SUBSTRATE - Synthetic nervous system layer
*
* Based on:
* - TinyTE (EMNLP 2025): Embedding-layer steering
* - DoRA (ICML 2024): Magnitude-direction decomposition
* - S-LoRA/Punica: Multi-adapter serving patterns
* - MMTEB: Multilingual embedding benchmarks
*/
export declare const NEURAL_CONSTANTS: {
readonly MAX_DRIFT_EVENTS: 1000;
readonly MAX_HISTORY_SIZE: 500;
readonly DEFAULT_DRIFT_THRESHOLD: 0.15;
readonly DEFAULT_DRIFT_WINDOW_MS: 60000;
readonly DRIFT_CRITICAL_MULTIPLIER: 2;
readonly VELOCITY_WINDOW_SIZE: 10;
readonly MAX_MEMORIES: 10000;
readonly MAX_CONTENT_LENGTH: 10000;
readonly MAX_ID_LENGTH: 256;
readonly DEFAULT_MEMORY_DECAY_RATE: 0.01;
readonly DEFAULT_INTERFERENCE_THRESHOLD: 0.8;
readonly DEFAULT_CONSOLIDATION_RATE: 0.1;
readonly MEMORY_FORGET_THRESHOLD: 0.01;
readonly CONSOLIDATION_SCORE_THRESHOLD: 0.5;
readonly MEMORY_CLEANUP_PERCENT: 0.1;
readonly RECALL_STRENGTH_BOOST: 0.1;
readonly MAX_TIME_JUMP_MINUTES: 1440;
readonly MAX_AGENTS: 1000;
readonly MAX_SPECIALTY_LENGTH: 100;
readonly AGENT_TIMEOUT_MS: 3600000;
readonly DEFAULT_AGENT_ENERGY: 1;
readonly TRAJECTORY_DAMPING: 0.1;
readonly MAX_TRAJECTORY_STEPS: 100;
readonly MAX_CLUSTER_AGENTS: 500;
readonly DEFAULT_CLUSTER_THRESHOLD: 0.7;
readonly DEFAULT_WINDOW_SIZE: 100;
readonly MIN_CALIBRATION_OBSERVATIONS: 10;
readonly STABILITY_WINDOW_SIZE: 10;
readonly ALIGNMENT_WINDOW_SIZE: 50;
readonly RECENT_OBSERVATIONS_SIZE: 20;
readonly DRIFT_WARNING_THRESHOLD: 0.3;
readonly STABILITY_WARNING_THRESHOLD: 0.5;
readonly ALIGNMENT_WARNING_THRESHOLD: 0.6;
readonly COHERENCE_WARNING_THRESHOLD: 0.5;
readonly EPSILON: 1e-8;
readonly ZERO_VECTOR_THRESHOLD: 1e-10;
readonly DEFAULT_DIMENSION: 384;
readonly DEFAULT_REFLEX_LATENCY_MS: 10;
};
export type LogLevel = 'debug' | 'info' | 'warn' | 'error';
export interface NeuralLogger {
log(level: LogLevel, message: string, data?: Record<string, unknown>): void;
}
/** Default console logger */
export declare const defaultLogger: NeuralLogger;
/** Silent logger for suppressing output */
export declare const silentLogger: NeuralLogger;
export interface DriftEvent {
readonly timestamp: number;
readonly magnitude: number;
readonly direction: Float32Array;
readonly category: 'normal' | 'warning' | 'critical';
readonly source?: string;
}
export interface NeuralMemoryEntry {
readonly id: string;
readonly embedding: Float32Array;
readonly content: string;
strength: number;
lastAccess: number;
accessCount: number;
consolidationLevel: number;
interference: number;
}
export interface AgentState {
readonly id: string;
position: Float32Array;
velocity: Float32Array;
attention: Float32Array;
energy: number;
mode: string;
lastUpdate: number;
}
export interface CoherenceReport {
readonly timestamp: number;
readonly overallScore: number;
readonly driftScore: number;
readonly stabilityScore: number;
readonly alignmentScore: number;
readonly anomalies: ReadonlyArray<{
readonly type: string;
readonly severity: number;
readonly description: string;
}>;
}
export interface NeuralConfig {
readonly dimension?: number;
readonly driftThreshold?: number;
readonly driftWindowMs?: number;
readonly memoryDecayRate?: number;
readonly interferenceThreshold?: number;
readonly consolidationRate?: number;
readonly reflexLatencyMs?: number;
readonly logger?: NeuralLogger;
}
/**
* Detects semantic drift and triggers reflexes based on embedding movement.
* Instead of asking "what is similar", asks "how far did we move".
*/
export declare class SemanticDriftDetector {
private baseline;
private history;
private driftEvents;
private config;
private logger;
private reflexes;
constructor(config?: NeuralConfig);
/**
* Set the baseline embedding (reference point)
*/
setBaseline(embedding: number[] | Float32Array): void;
/**
* Observe a new embedding and detect drift
*/
observe(embedding: number[] | Float32Array, source?: string): DriftEvent | null;
/**
* Calculate drift between two embeddings
*/
private calculateDrift;
/**
* Register a reflex callback for drift events
*/
registerReflex(name: string, callback: (event: DriftEvent) => void): void;
/**
* Trigger registered reflexes
*/
private triggerReflexes;
/**
* Get recent drift velocity (rate of change)
*/
getVelocity(): number;
/**
* Get drift statistics
*/
getStats(): {
currentDrift: number;
velocity: number;
criticalEvents: number;
warningEvents: number;
historySize: number;
};
/**
* Reset baseline to current position
*/
recenter(): void;
}
/**
* Implements hippocampal-like memory dynamics in embedding space.
* Memory strength decays, similar memories interfere, consolidation strengthens.
*/
export declare class MemoryPhysics {
private memories;
private config;
private lastUpdate;
private logger;
constructor(config?: NeuralConfig);
/**
* Encode a new memory
*/
encode(id: string, embedding: number[] | Float32Array, content: string): NeuralMemoryEntry;
/**
* Recall memories similar to a query (strengthens accessed memories)
*/
recall(query: number[] | Float32Array, k?: number): NeuralMemoryEntry[];
/**
* Apply time-based decay to all memories
*/
private applyDecay;
/**
* Consolidate memories (like sleep consolidation)
* Strengthens frequently accessed, weakly interfered memories
*/
consolidate(): {
consolidated: number;
forgotten: number;
};
/**
* Get memory statistics
*/
getStats(): {
totalMemories: number;
avgStrength: number;
avgConsolidation: number;
avgInterference: number;
};
private cosineSimilarity;
/**
* Force cleanup of weak memories when limit reached
*/
private forceCleanup;
}
/**
* Manages agent state as movement through embedding space.
* Decisions become geometric - no explicit state machine.
*/
export declare class EmbeddingStateMachine {
private agents;
private modeRegions;
private config;
private logger;
private lastCleanup;
constructor(config?: NeuralConfig);
/**
* Create or update an agent
*/
updateAgent(id: string, embedding: number[] | Float32Array): AgentState;
/**
* Remove stale agents that haven't been updated recently
*/
private cleanupStaleAgents;
/**
* Manually remove an agent
*/
removeAgent(id: string): boolean;
/**
* Define a mode region in embedding space
*/
defineMode(name: string, centroid: number[] | Float32Array, radius?: number): void;
/**
* Determine which mode an agent is in based on position
*/
private determineMode;
/**
* Get agent trajectory prediction
*/
predictTrajectory(id: string, steps?: number): Float32Array[];
/**
* Apply attention to agent state
*/
attendTo(agentId: string, focusEmbedding: number[] | Float32Array): void;
/**
* Get all agents in a specific mode
*/
getAgentsInMode(mode: string): AgentState[];
private euclideanDistance;
}
/**
* Enables multi-agent coordination through shared embedding space.
* Swarm behavior emerges from geometry, not protocol.
*/
export declare class SwarmCoordinator {
private agents;
private sharedContext;
private config;
private logger;
constructor(config?: NeuralConfig);
/**
* Register an agent with the swarm
*/
register(id: string, embedding: number[] | Float32Array, specialty?: string): void;
/**
* Update agent position (from their work/observations)
*/
update(id: string, embedding: number[] | Float32Array): void;
/**
* Update shared context (centroid of all agents)
*/
private updateSharedContext;
/**
* Get coordination signal for an agent (how to align with swarm)
*/
getCoordinationSignal(id: string): Float32Array;
/**
* Find agents working on similar things (for collaboration)
*/
findCollaborators(id: string, k?: number): Array<{
id: string;
similarity: number;
specialty: string;
}>;
/**
* Detect emergent clusters (specialization)
*/
detectClusters(threshold?: number): Map<string, string[]>;
/**
* Get swarm coherence (how aligned are agents)
*/
getCoherence(): number;
private cosineSimilarity;
/**
* Remove an agent from the swarm
*/
removeAgent(id: string): boolean;
}
/**
* Monitors system coherence via embedding patterns.
* Detects degradation, poisoning, misalignment before explicit failures.
*/
export declare class CoherenceMonitor {
private history;
private baselineDistribution;
private config;
private logger;
constructor(config?: NeuralConfig & {
windowSize?: number;
});
/**
* Record an observation
*/
observe(embedding: number[] | Float32Array, source?: string): void;
/**
* Establish baseline distribution
*/
calibrate(): void;
/**
* Generate coherence report
*/
report(): CoherenceReport;
private calculateDriftScore;
private calculateStabilityScore;
private calculateAlignmentScore;
private cosineSimilarity;
}
/**
* Unified neural embedding substrate combining all components.
* Acts like a synthetic nervous system with reflexes, memory, and coordination.
*/
export declare class NeuralSubstrate {
readonly drift: SemanticDriftDetector;
readonly memory: MemoryPhysics;
readonly state: EmbeddingStateMachine;
readonly swarm: SwarmCoordinator;
readonly coherence: CoherenceMonitor;
private config;
private logger;
private reflexLatency;
constructor(config?: NeuralConfig);
/**
* Process an embedding through the entire substrate
*/
process(embedding: number[] | Float32Array, options?: {
agentId?: string;
memoryId?: string;
content?: string;
source?: string;
}): {
drift: DriftEvent | null;
memory: NeuralMemoryEntry | null;
state: AgentState | null;
};
/**
* Query the substrate
*/
query(embedding: number[] | Float32Array, k?: number): {
memories: NeuralMemoryEntry[];
collaborators: Array<{
id: string;
similarity: number;
specialty: string;
}>;
coherence: CoherenceReport;
};
/**
* Get overall system health
*/
health(): {
driftStats: ReturnType<SemanticDriftDetector['getStats']>;
memoryStats: ReturnType<MemoryPhysics['getStats']>;
swarmCoherence: number;
coherenceReport: CoherenceReport;
};
/**
* Run consolidation (like "sleep")
*/
consolidate(): {
consolidated: number;
forgotten: number;
};
/**
* Calibrate coherence baseline
*/
calibrate(): void;
}
export default NeuralSubstrate;
//# sourceMappingURL=neural-embeddings.d.ts.map |