Spaces:
Paused
Paused
File size: 8,151 Bytes
529090e | 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 | /**
* Meta-Learning Engine
* Learns how to learn - optimizes learning strategies across tasks
*/
export interface LearningStrategy {
name: string;
parameters: Record<string, any>;
performance: {
tasksApplied: number;
avgImprovement: number;
bestDomain: string;
};
}
export interface TaskDomain {
name: string;
characteristics: string[];
optimalStrategy: string;
transferability: Map<string, number>; // How well learning transfers to other domains
}
export class MetaLearningEngine {
private strategies: Map<string, LearningStrategy> = new Map();
private domains: Map<string, TaskDomain> = new Map();
private learningHistory: Array<{
domain: string;
strategy: string;
improvement: number;
timestamp: Date;
}> = [];
constructor() {
this.initializeDefaultStrategies();
}
/**
* Initialize default learning strategies
*/
private initializeDefaultStrategies(): void {
const defaultStrategies: LearningStrategy[] = [
{
name: 'gradient_descent',
parameters: { learningRate: 0.01, momentum: 0.9 },
performance: { tasksApplied: 0, avgImprovement: 0, bestDomain: '' },
},
{
name: 'few_shot',
parameters: { examples: 5, temperature: 0.7 },
performance: { tasksApplied: 0, avgImprovement: 0, bestDomain: '' },
},
{
name: 'reinforcement',
parameters: { explorationRate: 0.1, discountFactor: 0.95 },
performance: { tasksApplied: 0, avgImprovement: 0, bestDomain: '' },
},
{
name: 'transfer_learning',
parameters: { sourceTask: '', fineTuneEpochs: 10 },
performance: { tasksApplied: 0, avgImprovement: 0, bestDomain: '' },
},
];
defaultStrategies.forEach(strategy => {
this.strategies.set(strategy.name, strategy);
});
}
/**
* Select optimal learning strategy for a task
*/
selectStrategy(domain: string, taskCharacteristics: string[]): LearningStrategy {
const domainInfo = this.domains.get(domain);
if (domainInfo && domainInfo.optimalStrategy) {
const strategy = this.strategies.get(domainInfo.optimalStrategy);
if (strategy) return strategy;
}
// Find strategy with best performance in similar domains
const strategies = Array.from(this.strategies.values());
const scored = strategies.map(strategy => {
const relevantHistory = this.learningHistory.filter(h => h.strategy === strategy.name);
const avgImprovement = relevantHistory.length > 0
? relevantHistory.reduce((sum, h) => sum + h.improvement, 0) / relevantHistory.length
: 0;
return { strategy, score: avgImprovement };
});
scored.sort((a, b) => b.score - a.score);
return scored[0]?.strategy || strategies[0];
}
/**
* Record learning outcome
*/
recordLearningOutcome(
domain: string,
strategy: string,
improvement: number
): void {
this.learningHistory.push({
domain,
strategy,
improvement,
timestamp: new Date(),
});
// Update strategy performance
const strategyObj = this.strategies.get(strategy);
if (strategyObj) {
strategyObj.performance.tasksApplied++;
strategyObj.performance.avgImprovement =
(strategyObj.performance.avgImprovement * (strategyObj.performance.tasksApplied - 1) + improvement) /
strategyObj.performance.tasksApplied;
}
// Update domain optimal strategy
this.updateDomainStrategy(domain);
}
/**
* Update optimal strategy for a domain
*/
private updateDomainStrategy(domain: string): void {
const domainHistory = this.learningHistory.filter(h => h.domain === domain);
if (domainHistory.length < 5) return; // Need enough data
// Group by strategy
const strategyPerformance = new Map<string, number[]>();
domainHistory.forEach(h => {
if (!strategyPerformance.has(h.strategy)) {
strategyPerformance.set(h.strategy, []);
}
strategyPerformance.get(h.strategy)!.push(h.improvement);
});
// Find best strategy
let bestStrategy = '';
let bestAvg = -Infinity;
strategyPerformance.forEach((improvements, strategy) => {
const avg = improvements.reduce((sum, val) => sum + val, 0) / improvements.length;
if (avg > bestAvg) {
bestAvg = avg;
bestStrategy = strategy;
}
});
// Update domain
if (!this.domains.has(domain)) {
this.domains.set(domain, {
name: domain,
characteristics: [],
optimalStrategy: bestStrategy,
transferability: new Map(),
});
} else {
this.domains.get(domain)!.optimalStrategy = bestStrategy;
}
}
/**
* Transfer learning from one domain to another
*/
async transferLearning(
sourceDomain: string,
targetDomain: string
): Promise<LearningStrategy | null> {
const source = this.domains.get(sourceDomain);
if (!source) return null;
// Check transferability
const transferScore = source.transferability.get(targetDomain) || 0;
if (transferScore > 0.5) {
// High transferability - use source domain's strategy
const strategy = this.strategies.get(source.optimalStrategy);
if (strategy) {
console.log(`📚 Transferring learning from ${sourceDomain} to ${targetDomain}`);
return strategy;
}
}
return null;
}
/**
* Optimize learning parameters
*/
optimizeParameters(strategyName: string): Record<string, any> {
const strategy = this.strategies.get(strategyName);
if (!strategy) return {};
// Simple parameter optimization based on historical performance
const recentHistory = this.learningHistory
.filter(h => h.strategy === strategyName)
.slice(-20);
if (recentHistory.length < 10) return strategy.parameters;
// Analyze if we should adjust learning rate (example)
const avgImprovement = recentHistory.reduce((sum, h) => sum + h.improvement, 0) / recentHistory.length;
if (avgImprovement < 0.1 && strategy.parameters.learningRate) {
// Low improvement - increase learning rate
strategy.parameters.learningRate *= 1.1;
} else if (avgImprovement > 0.5 && strategy.parameters.learningRate) {
// High improvement - decrease learning rate for fine-tuning
strategy.parameters.learningRate *= 0.9;
}
return strategy.parameters;
}
/**
* Get meta-learning statistics
*/
getStatistics(): {
totalStrategies: number;
totalDomains: number;
mostEffectiveStrategy: string;
avgImprovementRate: number;
} {
const strategies = Array.from(this.strategies.values());
const mostEffective = strategies.sort((a, b) =>
b.performance.avgImprovement - a.performance.avgImprovement
)[0];
const avgImprovement = this.learningHistory.length > 0
? this.learningHistory.reduce((sum, h) => sum + h.improvement, 0) / this.learningHistory.length
: 0;
return {
totalStrategies: this.strategies.size,
totalDomains: this.domains.size,
mostEffectiveStrategy: mostEffective?.name || 'none',
avgImprovementRate: avgImprovement,
};
}
}
export const metaLearningEngine = new MetaLearningEngine();
|