Paper 2026/213

Orbit: Optimizing Rescale and Bootstrap Placement with Integer Linear Programming Techniques for Secure Inference

William Seo, Carnegie Mellon University

Edward Chen, Carnegie Mellon University

Alex Ozdemir, Max Planck Institute for Security and Privacy

Fraser Brown, Carnegie Mellon University

Wenting Zheng, Carnegie Mellon University

Abstract

Fully Homomorphic Encryption (FHE) allows computation on encrypted data without decrypting it. In theory, FHE makes privacy-preserving machine learning possible. In practice, however, it remains impractically slow for real workloads. A major source of slowdown is bootstrap operations; in CKKS, a popular FHE scheme for tensor workloads, the slowdown is compounded by scale management and rescale operations. FHE compilers aim to make bootstrap placement and scale management efficient and easy by compiling high-level programs into low-level, optimized FHE computations. Unfortunately, existing approaches miss crucial optimization opportunities because they overlook a key property of CKKS programs: bootstrap and rescale placement are fundamentally coupled through the level budget. In this paper, we present Orbit, an FHE compiler that jointly optimizes bootstrap and rescale placement through a novel Integer Linear Programming (ILP) formulation that reasons about both ciphertext level and scale constraints. To make this formulation tractable for end-to-end programs, we introduce three techniques that reduce ILP complexity while preserving optimality. Across five convolutional neural networks and multiple cryptographic parameter configurations, Orbit achieves a geometric mean speedup up to 1.19× over DaCapo, 1.73× over Orion, and 1.52× over ReSBM, keeps compilation under 6 minutes, and retains model accuracy within 0.3% of plaintext execution.

@misc{cryptoeprint:2026/213,
      author = {Zikai Zhou and William Seo and Edward Chen and Alex Ozdemir and Fraser Brown and Wenting Zheng},
      title = {Orbit: Optimizing Rescale and Bootstrap Placement with Integer Linear Programming Techniques for Secure Inference},
      howpublished = {Cryptology {ePrint} Archive, Paper 2026/213},
      year = {2026},
      url = {https://eprint.iacr.org/2026/213}
}