Basic Block Agent

Concept

The Basic Block Agent is the low-level reinforcement-learning component in HiFuzz that fills basic blocks with instruction-level configurations. It is trained with PPO using a multi-head actor-critic design, combines extrinsic coverage feedback with semantic intrinsic novelty reward, and uses a dual-advantage actor loss to keep intrinsic and extrinsic learning signals separate.

First seen 7/10/2026

Last seen 7/10/2026

Evidence 6 chunks

Wiki v1

WIKI

Overview

The Basic Block Agent is the low-level agent in HiFuzz's two-level hierarchical reinforcement-learning architecture. While the high-level Program Agent selects the global structure of a test program, the Basic Block Agent decides the instructions inside each basic block (BB) at micro steps within that global context.[C1]

HiFuzz separates these roles because program-level choices such as memory layout, BB count, and control flow operate at a different abstraction level and timescale from instruction-level decisions such as operand selection, dependency management, and BB termination.[C1]

READ FULL ARTICLE →

NEIGHBORHOOD

No graph connections found for this entity yet. It may appear in future ingestion runs.

explore full graph →

RELATIONSHIPS

8 connections

HiFuzz ← uses 100% 2e

HiFuzz uses a Basic Block Agent as the low-level executor for instruction filling.

Proximal Policy Optimization uses → 100% 2e

The Basic Block Agent is trained with PPO under a multi-head Actor-Critic architecture.

Intrinsic Novelty Reward uses → 100% 2e

The Basic Block Agent receives intrinsic novelty rewards from the BB encoder.

Dual-Advantage Actor Loss uses → 90% 2e

The Basic Block Agent employs a dual-advantage actor loss combining intrinsic and extrinsic advantages.

Multi-Head Actor-Critic uses → 100% 2e

The Basic Block Agent uses a multi-head Actor-Critic architecture with separate value heads for intrinsic and extrinsic returns.

HiFuzz part of → 100% 2e

The Basic Block Agent is a component of the HiFuzz dual-agent architecture.

Dual-Advantage Actor Loss ← part of 90% 1e

The dual-advantage actor loss is a design contribution within the Basic Block Agent's training objective.

Multi-Head Actor-Critic ← part of 100% 1e

The multi-head Actor-Critic is the architecture used by the Basic Block Agent.

LINKED ENTITIES

5 links

HiFuzz USES Extracted graph relationship

Proximal Policy Optimization USES Extracted graph relationship

Intrinsic Novelty Reward USES Extracted graph relationship

Dual-Advantage Actor Loss USES Extracted graph relationship

Multi-Head Actor-Critic USES Extracted graph relationship

CITATIONS

7 sources

7 citations — click to expand

[1] The Basic Block Agent is the low-level agent in HiFuzz's two-level hierarchical RL architecture and decides instructions inside each basic block. HiFuzz: Hierarchical Reinforcement Learning for Semantic-Aware and Adaptive CPU Fuzzing

[2] HiFuzz's BB Level Config specifies instruction-category mix and termination mode, while a constrained generator instantiates concrete instructions under semantic and legality constraints. HiFuzz: Hierarchical Reinforcement Learning for Semantic-Aware and Adaptive CPU Fuzzing

[3] The Basic Block Agent is trained with PPO under a multi-head Actor-Critic architecture; its actor emits BB-level configuration and its critic has separate intrinsic and extrinsic value heads. HiFuzz: Hierarchical Reinforcement Learning for Semantic-Aware and Adaptive CPU Fuzzing

[4] The Basic Block Agent uses an intrinsic novelty reward computed as cluster-distance novelty on frozen Semantic-Aware Basic Block Encoder embeddings, and its extrinsic reward is based on coverage improvement normalized by a dynamic baseline. HiFuzz: Hierarchical Reinforcement Learning for Semantic-Aware and Adaptive CPU Fuzzing

[5] The Basic Block Agent uses a dual-advantage actor loss and a total objective with per-head critic MSE losses plus an entropy bonus. HiFuzz: Hierarchical Reinforcement Learning for Semantic-Aware and Adaptive CPU Fuzzing

[6] HiFuzz implementation trains the Basic Block Agent with PPO on top of a multi-head Actor-Critic using α_int = 1, α_ext = 2, and γ = 0.75. HiFuzz: Hierarchical Reinforcement Learning for Semantic-Aware and Adaptive CPU Fuzzing

[7] The Semantic-Aware Basic Block Encoder drives an online cluster-distance novelty estimator whose output is the intrinsic reward used by the Basic Block Agent. HiFuzz: Hierarchical Reinforcement Learning for Semantic-Aware and Adaptive CPU Fuzzing