Interpreting Neural Branch Predictors: An Attention-Based Approach for Explainable Branch Prediction

References

[1] J. E. Smith, “A study of branch prediction strategies,” in 25 Years of the International Symposia on Computer Architecture (Selected Papers), pp. 202–215, 1998, doi: 10.1145/285930.285980.

[2] T.-Y. Yeh and Y. N. Patt, “Two-Level Adaptive Training Branch Prediction,” in Proc. 24th Annu. Int. Symp. Microarchitecture (MICRO), 1991, pp. 51–61, doi: 10.1145/123465.123475. ACM

[3] S. McFarling, Combining Branch Predictors, Technical Report WRL TN-36. Digital Equipment Corporation, 1993.

[4] D. A. Jiménez and C. Lin, “Dynamic branch prediction with perceptrons,” in Proc. 7th Int. Symp. High-Performance Computer Architecture, 2001, pp. 143–154, doi: 10.1109/HPCA.2001.903263. Crossref

[5] A. Seznec, “A new case for the TAGE branch predictor,” in Proc. 44th Annu. IEEE/ACM Int. Symp. Microarchitecture, 2011, pp. 117–127, doi: 10.1145/2155620.2155635. ACM

[6] S. Atukorala, “Branch prediction methods used in modern superscalar processors,” in Proc. ICICS 1997 Int. Conf. Information, Communications and Signal Processing, vol. 3, pp. 1475–1479, 1997, doi: 10.1109/ICICS.1997.652237. Crossref

[7] R. Parihar, “Branch Prediction Techniques and Optimizations,” 2009.

[8] C. M. Bhamare and M. Bhamare, “Predictive Branching Methods and Architectures–A survey,” 2011.

[9] S. Mittal, “A survey of techniques for dynamic branch prediction,” Concurrency and Computation: Practice and Experience, vol. 31, 2018.

[10] L. Vintan, “NEURAL BRANCH PREDICTION: FROM THE FIRST IDEAS, TO IMPLEMENTATIONS IN ADVANCED MICROPROCESSORS AND MEDICAL APPLICATIONS,” 2019.

[11] M. Eggen, J. Lysnæs-Larsen, and I. Strümke, “Integrating attention into explanation frameworks for language and vision transformers,” arXiv preprint arXiv:2508.08966, 2025, doi: 10.48550/arXiv.2508.08966. arXiv

[12] B. Wen, K. P. Subbalakshmi, and F. Yang, “Revisiting Attention Weights as Explanations from an Information Theoretic Perspective,” arXiv preprint arXiv:2211.07714, 2022, doi: 10.48550/arXiv.2211.07714. arXiv

[13] J. Bastings and K. Filippova, “The elephant in the interpretability room: Why use attention as explanation when we have saliency methods?,” in Proc. Third BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP, 2020, pp. 149–155, doi: 10.18653/v1/2020.blackboxnlp-1.14. ACL Anthology

[14] A. K. Mohankumar, P. Nema, S. Narasimhan, M. M. Khapra, B. V. Srinivasan, and B. Ravindran, “Towards Transparent and Explainable Attention Models,” in Proc. 58th Annu. Meeting Association for Computational Linguistics, 2020, pp. 4206–4216, doi: 10.18653/v1/2020.acl-main.387. ACL Anthology

[15] L. Hu, X. Wang, Y. Liu, N. Liu, M. Huai, L. Sun, and D. Wang, “Towards Stable and Explainable Attention Mechanisms,” IEEE Transactions on Knowledge and Data Engineering, vol. 37, pp. 3047–3061, 2025, doi: 10.1109/TKDE.2025.3538583.

[16] S. Jain and B. C. Wallace, “Attention is not explanation,” in Proc. 2019 Conf. North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), 2019, pp. 3543–3556, doi: 10.18653/v1/N19-1357. ACL Anthology

[17] S. Wiegreffe and Y. Pinter, “Attention is not not explanation,” in Proc. 2019 Conf. Empirical Methods in Natural Language Processing and 9th Int. Joint Conf. Natural Language Processing (EMNLP-IJCNLP), 2019, pp. 11–20, doi: 10.18653/v1/D19-1002. ACL Anthology

[18] C. Grimsley, E. Mayfield, and J. R. S. Bursten, “Why Attention is Not Explanation: Surgical Intervention and Causal Reasoning about Neural Models,” in Proc. Twelfth Language Resources and Evaluation Conf., 2020, pp. 1780–1790. ACL Anthology

[19] A. R. Akula and S.-C. Zhu, “Attention cannot be an Explanation,” arXiv preprint arXiv:2201.11194, 2022, doi: 10.48550/arXiv.2201.11194. arXiv

[20] G. A. Mihaila, “Learning to Explain: Supervised Token Attribution from Transformer Attention Patterns,” arXiv preprint arXiv:2601.14112, 2026, doi: 10.48550/arXiv.2601.14112. arXiv

[21] M. P. Ayyar, J. Benois-Pineau, and A. Zemmari, “There is More to Attention: Statistical Filtering Enhances Explanations in Vision Transformers,” arXiv preprint arXiv:2510.06070, 2025, doi: 10.48550/arXiv.2510.06070. arXiv

[22] M. V. Ntrougkas, N. Gkalelis, and V. Mezaris, “T-TAME: Trainable Attention Mechanism for Explaining Convolutional Networks and Vision Transformers,” IEEE Access, vol. 12, pp. 76880–76900, 2024, doi: 10.1109/ACCESS.2024.3405788. Crossref

[23] G. Liu, J. Zhang, A. B. Chan, and J. Hsiao, “Human Attention-Guided Explainable Artificial Intelligence for Computer Vision Models,” Neural networks : the official journal of the International Neural Network Society, vol. 177, Art. no. 106392, 2023.

[24] R. Qi, Y. Zheng, Y. Yang, C. C. Cao, and J. Hsiao, “Explanation strategies in humans versus current explainable artificial intelligence: Insights from image classification,” British Journal of Psychology, vol. 117, pp. 479–502, 2024.

[25] P. Srinivasu, N. Sandhya, R. Jhaveri, and R. Raut, “From Blackbox to Explainable AI in Healthcare: Existing Tools and Case Studies,” Mobile Information Systems, 2022, doi: 10.1155/2022/8167821. Wiley

[26] Y. Xu, M. D. Plumbley, and W. Wang, “Explainable AI in Speaker Recognition–Attention Map Visualisation and Evaluation,” arXiv preprint arXiv:2606.22901, 2026, doi: 10.48550/arXiv.2606.22901. arXiv

[27] Y. Zhu, “Applications of Attention Mechanisms in Explainable Machine Learning,” Academic Journal of Computing & Information Science, 2025, doi: 10.25236/AJCIS.2025.081008. Academic Journal of Computing & Information Science

[28] C. Bai, J. Huang, X. Wei, Y., S. Li, H. Zheng, B. Yu, and Y. Xie, “ArchExplorer: Microarchitecture Exploration Via Bottleneck Analysis,” in Proc. 2023 56th IEEE/ACM Int. Symp. Microarchitecture (MICRO), pp. 268–282, 2023.

[29] A. Dutilleul, H. Pompougnac, N. Derumigny, G. Rodríguez, V. Trophime, C. Guillon, and F. Rastello, “Performance Debugging through Microarchitectural Sensitivity and Causality Analysis,” arXiv preprint arXiv:2412.13207, 2024, doi: 10.48550/arXiv.2412.13207. arXiv

[30] H. Alkhiri, S. Kumar, H. Alsolai, H. Dafaalla, S. Asklany, O. Alrusaini, A. Alqazzaz, and M. Alshammeri, “Enhancing micro-architecture optimization using explainable fuzzy neural networks with multi-fidelity reinforcement learning,” PeerJ Computer Science, vol. 12, Art. no. e3429, 2026, doi: 10.7717/peerj-cs.3429. PeerJ

[31] A. P. Kuruvila, X. Meng, S. Kundu, G. Pandey, and K. Basu, “Explainable Machine Learning for Intrusion Detection via Hardware Performance Counters,” IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 41, pp. 4952–4964, 2022, doi: 10.1109/TCAD.2022.3149745. Crossref

[32] Y. Nasser and M. Nassar, “Toward Hardware-Assisted Malware Detection Utilizing Explainable Machine Learning: A Survey,” IEEE Access, vol. 11, pp. 131273–131288, 2023, doi: 10.1109/ACCESS.2023.3335187. Crossref

[33] R. Gupta, A. Jain, A. Gonzalez, A. Novikov, P.-S. Huang, M. Balog, M. Eisenberger, S. Shirobokov, N. Vu, M. Dixon, B. Nikolić, P. Ranganathan, and S. Karandikar, “ArchAgent: Agentic AI-driven Computer Architecture Discovery,” arXiv preprint arXiv:2602.22425, 2026, doi: 10.48550/arXiv.2602.22425. arXiv

[34] H. Abdelkhalik, Y. Arafa, N. Santhi, and A.-H. A. Badawy, “Demystifying the Nvidia Ampere Architecture through Microbenchmarking and Instruction-level Analysis,” in Proc. 2022 IEEE High Performance Extreme Computing Conf. (HPEC), 2022, pp. 1–8, doi: 10.1109/HPEC55821.2022.9926299. Crossref

[35] J. Gómez-Luna, I. E. Hajj, I. Fernandez, C. Giannoula, G. F. Oliveira, and O. Mutlu, “Benchmarking a New Paradigm: Experimental Analysis and Characterization of a Real Processing-in-Memory System,” IEEE Access, vol. PP, pp. 1–1, 2022, doi: 10.1109/ACCESS.2022.3174101.

[36] D. Große, S. Reitinger, and L. Klemmer, “SonicRV: An Educational Platform for Web-Based Simulation and Visualization of RISC-V Processor Architectures,” IEEE Access, vol. 14, pp. 13849–13864, 2026, doi: 10.1109/ACCESS.2026.3656022. Crossref

[37] S. Prakash, A. Cheng, A. Tschand, M. Mazumder, V. Gohil, J., J. Yik, Z. Wan, J. Quaye, E. L. Alvanaki, A. Kumar, C. Mazumdar, T. Khare, A. Ingare, I. Uchendu, R. Ghosal, A. Tyagi, C. Wang, A. M. Garavagno, et al., “QuArch: A Benchmark for Evaluating LLM Reasoning in Computer Architecture,” arXiv preprint arXiv:2510.22087, 2025, doi: 10.48550/arXiv.2510.22087.

[38] R. Gacitúa, J. Pereira, and M. Klafft, “Explainability in Software Architectural Decisions: The ADR-E Framework and Empirical Evaluation,” IEEE Access, vol. 14, pp. 9038–9061, 2026, doi: 10.1109/ACCESS.2025.3648573.

[39] A. Hanif, R. Shawi, A. Beheshti, and B. Benatallah, “Survey on Explainable AI for Traditional Machine Learning and Domains,” ACM Computing Surveys, vol. 58, pp. 1–42, 2026, doi: 10.1145/3806829. ACM

[40] Z. Carmichael, T. Moon, and S. A. Jacobs, “Learning Interpretable Models Through Multi-Objective Neural Architecture Search,” arXiv preprint arXiv:2112.08645, 2021, doi: 10.48550/arXiv.2112.08645. arXiv

[41] E. O. Nonum, O. Avwokuruaye, and A. M. Umar, “Social Engineering: Understanding Human Factors in Cyber Security,” International Journal of Convergent and Informatics Science Research, vol. 7, no. 9, 2025, doi: 10.70382/hijcisr.v07i9.032. International Journal of Convergent and Informatics Science Research

[42] E. O. Nonum, O. Avwokuruaye, and T. M. Ezemonye, “Role of Open Source Intelligence (OSINT) in Cybersecurity and Threat Analysis,” International Journal of Latest Technology in Engineering, Management & Applied Science, vol. XIV, no. III, pp. 189–200, 2025, doi: 10.51583/IJLTEMAS.2025.140300023. IJLTEMAS

@article{1723172,
      author = {Otene P. U., Nonum E. O.},
      title = {Interpreting Neural Branch Predictors: An Attention-Based Approach for Explainable Branch Prediction},
      journal = {Iconic Research And Engineering Journals},
      year = {2026},
      volume = {10},
      number = {3},
      pages = {2616-2628},
      issn = {2456-8880},
      url = {https://www.irejournals.com/formatedpaper/1723172.pdf},
      abstract = {Neural branch predictors, from perceptron-based designs to modern learned architectures, achieve prediction accuracy that classical table-based mechanisms cannot match, but they do so by replacing a human-readable counter with an opaque weight vector. We propose an attention-based branch predictor (ABP) that operates over a fixed-length branch-history window and produces, alongside every taken/not-taken prediction, a per-prediction attention distribution over the history positions that informed it. We evaluate ABP against a 2-bit saturating-counter baseline and a perceptron predictor on a controlled synthetic branch trace containing three canonical branch families (loop-exit, data-dependent, and recursive-call branches), and introduce a perturbation-based faithfulness test that measures whether the prediction changes when the most attention-weighted history bit is flipped, compared against flipping a bit chosen at random or a bit chosen by input-gradient saliency. On this trace, ABP matches the 2-bit counter overall (68.4% vs. 68.3% accuracy) and outperforms the perceptron baseline (65.8%), with the largest gain on the data-dependent branch family and the largest shortfall on the recursive-call family. The faithfulness test shows both attention-guided and gradient-guided perturbations roughly double the flip rate of random perturbation (9.4% and 8.8% vs. 4.4%), and the two explanation methods agree on the most important history position in 46.4% of cases, far above the 3.1% expected by chance. An analytical overhead estimate shows that exposing this explanation costs approximately 2,100× more multiply-accumulate operations per prediction than the perceptron baseline, making explicit a real hardware cost that prior interpretability discussions in this space have left unquantified. We report these results as a controlled proof-of-concept on synthetic data rather than a production-scale evaluation, and discuss what would be required to extend the study to real hardware traces.},
      keywords = {Explainable AI, Branch prediction, Computer architecture, Attention mechanisms, Interpretability, Neural predictors},
      month = {September},
  }