/**
 * Agentic QE v3 - Neural Topology Optimizer
 * ADR-034: RL-based swarm topology optimization
 *
 * Implements reinforcement learning-based topology optimization:
 * - Q-learning with value network for state evaluation
 * - Experience replay for stable training
 * - Epsilon-greedy exploration
 * - Multi-objective reward (min-cut, efficiency, load balance)
 */
import type { SwarmTopology, TopologyOptimizerConfig, OptimizationResult, OptimizationStats, ExportedModel } from './types';
/**
 * Neural Topology Optimizer using reinforcement learning
 *
 * Uses a value network to estimate state values and learns to select
 * topology modifications that improve overall swarm performance.
 */
export declare class NeuralTopologyOptimizer {
    /** Primary value network */
    private valueNetwork;
    /** Target network for stable learning */
    private targetNetwork;
    /** Experience replay buffer */
    private replayBuffer;
    /** Optimizer configuration */
    private config;
    /** Reference to topology being optimized */
    private topology;
    /** Previous state for learning */
    private prevState;
    /** Previous min-cut estimate */
    private prevMinCut;
    /** Current exploration rate */
    private epsilon;
    /** Simulation time */
    private time;
    /** Total optimization steps */
    private totalSteps;
    /** Episode count */
    private episodes;
    /** Cumulative reward */
    private cumulativeReward;
    /** Action counts */
    private actionCounts;
    /** Min-cut history */
    private minCutHistory;
    /** Reward history */
    private rewardHistory;
    /** Last action taken (for feedback) */
    private lastAction;
    constructor(topology: SwarmTopology, config?: Partial<TopologyOptimizerConfig>);
    /**
     * Run one optimization step
     */
    optimizeStep(): OptimizationResult;
    /**
     * Run multiple optimization steps
     */
    optimize(steps: number): OptimizationResult[];
    /**
     * Provide external feedback (e.g., from task completion)
     */
    provideFeedback(reward: number): void;
    /**
     * Get skip regions (low activity areas)
     */
    getSkipRegions(): string[];
    /**
     * Get optimization statistics
     */
    getStats(): OptimizationStats;
    /**
     * Reset optimizer state
     */
    reset(): void;
    /**
     * Hard reset (clear learning)
     */
    hardReset(): void;
    /**
     * Export learned model
     */
    exportModel(): ExportedModel;
    /**
     * Import learned model
     */
    importModel(model: ExportedModel): void;
    /**
     * Extract features from topology for state representation
     */
    private extractFeatures;
    /**
     * Select action using epsilon-greedy policy
     */
    private selectAction;
    /**
     * Convert index to action
     */
    private indexToAction;
    /**
     * Generate random action
     */
    private randomAction;
    /**
     * Simulate action effect on state (for lookahead)
     */
    private simulateAction;
    /**
     * Apply action to topology
     */
    private applyAction;
    /**
     * Calculate multi-objective reward
     */
    private calculateReward;
    /**
     * Train from replay buffer
     */
    private trainFromReplay;
    private estimateMinCut;
    private hasConnection;
    private getDensity;
    private getAverageDegree;
    private getMinDegree;
    private getAverageWeight;
    private getWeightVariance;
    private getAverageLatency;
    private measureCommunicationEfficiency;
    private getClusteringCoefficient;
    private getNeighbors;
    private getLoadStats;
}
/**
 * Create a neural topology optimizer
 */
export declare function createNeuralTopologyOptimizer(topology: SwarmTopology, config?: Partial<TopologyOptimizerConfig>): NeuralTopologyOptimizer;
//# sourceMappingURL=topology-optimizer.d.ts.map