Skip to content

COMPARISON OF FULLY HOMOMORPHIC ENCRYPTION AND GARBLED CIRCUITS APPROACHES IN PRIVACY-PRESERVING MACHINE LEARNING Kalyan Cheerla Thesis Prepared for the Degree of MASTER OF SCIENCE UNIVERSITY OF NORTH TEXAS July 2025

Year: 2025

type: thesis collection: ppml status: auto

Summary

Auto-extracted summary (unreviewed)

This master's thesis (University of North Texas, 2025) compares FHE and garbled circuits for secure two-party inference of a two-layer neural network, using CKKS from Microsoft SEAL and the TinyGarble2.0 framework, and evaluates output error, round-trip time, peak memory, communication overhead and rounds under a semi-honest threat model. It concludes that the modular GC approach is faster and uses less memory while FHE supports non-interactive inference; the excerpt is mostly front matter and table of contents without numeric results.

Extracted details

Field Value Source Confidence
contribution comparative evaluation of FHE (CKKS/SEAL) and GC (TinyGarble2.0) for secure neural network inference llm 0.75
contribution reproducible implementations of both protocols llm 0.75
contribution analysis of round-trip time, memory, communication, interaction rounds and output deviation llm 0.75
schemes CKKS llm 0.75
schemes garbled circuits llm 0.75
schemes oblivious transfer llm 0.75
libraries SEAL llm 0.75
libraries TinyGarble2.0 llm 0.75
libraries EMP-tool llm 0.75
models two-layer neural network llm 0.75
tasks secure neural network inference llm 0.75
security level 128-bit llm 0.75
threat model semi-honest two-party llm 0.75
interaction other llm 0.75
code url https://github.com/microsoft/SEAL/blob/main/native/src/seal/util/globals.cpp (URL found in the full text) regex 0.50
code url https://github.com/IntelLabs/emp-tool (URL found in the full text) regex 0.50
code url https://github.com/Microsoft/SEAL (URL found in the full text) regex 0.50
code url https://github.com/kalyancheerla/snni-fhe-gc llm 0.75
llm relevance low llm 0.75

Keyword mentions

Counts of vocabulary terms in the full text (first page of occurrence in brackets).

  • scheme: Garbled circuits 61 [p.1], CKKS 36 [p.2], MPC 26 [p.2], Oblivious transfer 13 [p.5], Leveled / somewhat HE 6 [p.4], TEE / confidential computing 5 [p.36], Partially homomorphic encryption 5 [p.4], BFV 3 [p.26], Secret sharing 3 [p.32], Paillier 2 [p.25]
  • library: Microsoft SEAL 19 [p.2], EVA / CHET 1 [p.32], ABY / ABY3 1 [p.32], CrypTen 1 [p.33]
  • model: Transformer 4 [p.5], GPT 2 [p.12], BERT 1 [p.33], LLaMA 1 [p.79], Logistic regression 1 [p.83]
  • dataset: Breast cancer (Wisconsin) 1 [p.79], Credit card / German credit 1 [p.79]
  • technique: ReLU 43 [p.11], Sigmoid 31 [p.11], Rescaling 23 [p.5], Non-interactive 23 [p.2], Polynomial approximation 18 [p.5], Two-party computation 12 [p.23], Semi-honest model 12 [p.2], Matrix multiplication 8 [p.28], Relinearization 6 [p.39], Bootstrapping 5 [p.26], Modulus switching 5 [p.26], SIMD packing 3 [p.62], Rotations 3 [p.48], Lookup table 3 [p.49], Malicious model 3 [p.37], Compiler 3 [p.58], NTT 2 [p.26]
  • assumption: 128-bit security 2 [p.10]
  • task: Language modeling / generation 8 [p.41], Medical / healthcare 7 [p.2], Finance 7 [p.2], Genomics 4 [p.19]
  • topic: Homomorphic encryption 146 [p.2], Private / secure inference 44 [p.4], Privacy-preserving machine learning 20 [p.1], Benchmark 8 [p.19], Open source 6 [p.47], Large language models 2 [p.12]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/ppml/theses-and-textbooks/cheerla-2025_comparison-of-fhe-and-garbled-circuits-approaches-in-ppml-msc-thesis.pdf ppml/theses-and-textbooks 85 698.1 KB afb0fb654004 was Cheerla - 2025 - Comparison of Fully Homomorphic Encryption and Garbled Circuits approaches in Privacy-Preserving Mac.pdf

Database key: comparison-of-fully-homomorphic-encryption-and-garbled-circuits · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set comparison-of-fully-homomorphic-encryption-and-garbled-circuits COLUMN VALUE