Encrypted inferences over decision trees

Encrypted inferences over decision trees

by Alex Shpurov, Daniel Johnson

Formal & Physical Sciences AI — Models & Research UYQArtificial intelligence (AI)UYQMMachine learning
🎙 Alex Shpurov, Daniel Johnson 👥 5K 📅 August 11, 2026 ⏱ 21 min 👁 11 📄 expert opinion 🧭 2026-08-15
Available in: English (current) Français

Keywords

FHEdecision treesencrypted inferenceprivacypost-quantum

Summary

The talk, presented by Alex Shpurov and Daniel Johnson from 01 Quantum, introduces the concept of encrypted inference over decision trees using Fully Homomorphic Encryption (FHE). The first part, by Shpurov, demonstrates the 01 Quantum AI Marketplace, a platform that allows data owners to encrypt their data and model owners to encrypt their models, enabling inference without revealing raw data or model parameters. The demo shows a credit card fraud detection scenario where a client encrypts data, sends it to a model owner, and receives encrypted results that only the client can decrypt. The second part, by Johnson, explains the technical challenges of FHE development, including the need to replace if-else branching with polynomial approximations, the use of vector rotations, and the importance of managing multiplicative depth and precision. The talk concludes with a Q&A session discussing performance, scalability, and future directions such as neural networks and LLMs.

148 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides a practical demonstration of FHE-based encrypted inference, which is valuable for understanding the feasibility and workflow. The argumentation is coherent, explaining the need for FHE in privacy-preserving AI and the challenges of implementing decision trees in an encrypted domain. However, the technical depth is limited, and the presentation is more of an overview than a rigorous scientific exposition. The speakers rely on their expertise and the demo rather than formal proofs or detailed comparisons.

Scientific Rigor, Source Quality, Title Accuracy

The talk does not cite specific sources or references. The description mentions the speakers’ backgrounds and the open-source nature of the platform, but no external links are provided. The title accurately reflects the content. The presentation is based on the speakers’ own work and experience, which adds credibility but lacks external validation. The lack of citations reduces the scientific rigor.

152 words

Title / Content Match

The title accurately reflects the content, which focuses on encrypted inference over decision trees using FHE.

Quality & Reliability

7/10

The talk is given by practitioners with relevant expertise in FHE and cryptography, and includes a live demo. However, it is largely a high-level overview with limited technical depth and no formal citations or peer-reviewed references.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Contribution & Novelties

The talk presents a novel platform for encrypted inference over decision trees, demonstrating a practical implementation. It highlights the engineering challenges of FHE, such as polynomial approximations and packing strategies. The speakers also discuss future work on neural networks and LLMs, indicating ongoing research.

Pour aller plus loin :

89 words

Radar Profile

The radar profile shows balanced scores across information quantity, quality, technical level, and reliability, with a slight emphasis on quantity and quality. This indicates a well-rounded presentation that is informative and credible, though not deeply technical.

Reliability 7/10