Hybrid Framework iResolveX Boosts Precision in Indirect Call Resolution

Global: Hybrid Framework iResolveX Boosts Precision in Indirect Call Resolution

A research team has introduced iResolveX, a hybrid analysis system designed to improve the identification of indirect call targets in stripped or optimized binaries. The framework was described in a preprint posted to arXiv on January 2026, and it aims to balance recall and precision for reverse‑engineering tasks.

Challenges in Indirect Call Resolution

Indirect call resolution remains a persistent obstacle for analysts constructing control‑flow graphs, particularly when source code is unavailable or compiler optimizations obscure call sites. Traditional static analyses tend to over‑approximate, generating many false positives, while purely machine‑learning‑based methods may miss legitimate targets.

Hybrid Multi‑Layered Approach

iResolveX addresses these trade‑offs through a three‑layer pipeline. The first layer employs a conservative value‑set analysis (BPA) to capture the majority of possible targets, ensuring high recall. The second layer introduces a learning‑based soft‑signature scorer, iScoreGen, which assigns confidence scores to candidate edges. The final layer, iScoreRefine, performs selective inter‑procedural backward analysis with memory inspection to prune unlikely targets further.

Evaluation and Results

Testing on SPEC CPU2006 benchmarks and a collection of real‑world binaries, iScoreGen alone reduced the number of predicted indirect call targets by 19.2 % on average while preserving a recall rate of 98.2 %. When combined with iScoreRefine, the overall reduction reached 44.3 % with a recall of 97.8 %, representing a modest 0.4 % drop from the baseline BPA recall.

Potential Impact

The authors note that iResolveX can operate in both a conservative mode that prioritizes recall and an F1‑optimized configuration that emphasizes overall accuracy. By annotating indirect edges with confidence scores, the system enables downstream analyses—such as vulnerability discovery or binary similarity assessment—to select the precision level that best fits their objectives.

This report is based on information from arXiv, licensed under Academic Preprint / Open Access. Based on the abstract of the research paper. Full text available via ArXiv.