# Juan de la Cierva fellowship — Antonio Guimarães Reference JDC2024-055789-I (call 2024; applied Jan 2025, awarded Oct 2025). Fellowship: Jan 1, 2026 – Dec 31, 2027 (two years). Host: IMDEA Software Institute, Madrid, Spain. Supervisor: Ignacio Cascudo. Grant JDC2024-055789-I, funded by MICIU/AEI/10.13039/501100011033 and by ESF+. ## The programme The Juan de la Cierva programme is a postdoctoral fellowship funded by the Spanish Ministry of Science, Innovation and Universities through the State Research Agency (AEI). Together with the more senior Ramón y Cajal programme, it is regarded as one of the two most prestigious nationally funded fellowships for a research career in Spain. It was created in 2004 and has had a call every year since. It is named after Juan de la Cierva (1895–1936), the Spanish engineer who invented the autogiro and the articulated rotor, which made the first stable flight of a rotary-wing aircraft possible. See more on Wikipedia: https://en.wikipedia.org/wiki/Juan_de_la_Cierva_Scholarship ## The proposal: Secure and efficient fully homomorphic encryption for privacy-preserving computation Keywords: cryptography, privacy, privacy-preserving computation, fully homomorphic encryption, verifiable computation, efficient cryptographic software. In the processing of sensitive information, privacy has always been an aspect of utmost interest. The popularization of large-scale data processing applications brought renewed interest to the topic as those often ought to be employed in a privacy-preserving manner. Additionally, as these applications are commonly outsourced to third-party cloud providers, new security threats arise from vulnerabilities intrinsic to the public cloud computing scenario. Cryptography provides important tools for enabling formal guarantees of privacy, and it is broadly used to protect data at rest and in transit. On the other hand, data might be equally vulnerable during processing while protecting it is generally seen as a major challenge. In this context, Fully Homomorphic Encryption (FHE) appears as an ideal solution by enabling computation to be performed directly over encrypted data without decryption. Practical performance is generally the main impairment for the popularization of FHE, as initial FHE constructions introduced a large performance overhead even on the most basic applications. Notwithstanding, significant progress has been consistently made in the last decade with new FHE schemes providing practical solutions for a wide range of applications. Among the many examples of modern FHE use cases, the homomorphic evaluation of neural network inference algorithms stands out by its promising data processing capabilities and the multitude of efficient FHE methods available for evaluating it. Further progress is nonetheless still necessary on many fronts, as FHE can only be deemed fully practical for applications in somewhat restricted scenarios. For example, the evaluation of very high-precision non-algebraic functions is a challenge that FHE has yet to overcome. Other security concerns also arise when considering the broader employment of FHE, especially against malicious adversaries. While FHE provides strong guarantees of confidentiality, it is inherently malleable, allowing malicious parties to modify computation results and perform undesired operations over the encrypted data. Protecting against such adversaries is fundamental in many real-world scenarios, and Verifiable Computation (VC) techniques are the ideal solution for it. The challenges of using these techniques in conjunction lie primarily in the difficulties of setting up compatible parameters and computation models, as well as in dealing with the combined performance overhead introduced by these techniques. On the other hand, the vast number of existing constructions for VC and FHE creates many possibilities for practical protocols, both for generic computing and specialized applications. Research goals: 1. To advance the efficiency of fully homomorphic encryption, that allows for its practical applicability in more scenarios such as the evaluation of high-precision non-algebraic functions mentioned above. 2. To provide more efficient verifiable computation protocols that ensure correct computations over fully homomorphic encrypted data when this computation is performed by an untrusted party. ## Results so far Work supported by grant JDC2024-055789-I. - [TUTTI: Parallel CKKS Bootstrapping in under 3 ms on CPU](https://antonioguimaraes.org/publications/tutti.md) (arXiv:2611.12854), goal 1: CKKS bootstrapping with lower latency than TFHE: a single message is bootstrapped in 4.6 ms on 32 threads, against 7.3 ms for a 4-bit TFHE-rs bootstrapping on the same machine (2.4 ms against 4.5 ms on a laptop), and 4096 messages in 46 to 83.4 ms on 192 threads, up to 3 times faster than the previous low-latency approach (SHIP). - [Tulva: Practical Asynchronous Threshold FHE](https://antonioguimaraes.org/publications/tulva.md) (IEEE S&P 2027), goal 1: Practical threshold FHE, which removes the single point of failure of FHE (one secret key that can read everything) at a concrete cost: any t of K parties decrypt asynchronously, without coordinating, in 3.7 ms per partial decryption for a 16-of-512 threshold, and the base scheme's ciphertexts are left unchanged. Security holds under the standard LWE assumption even given partial decryptions of challenge ciphertexts. - [Batch, Pack, and Prove: More Efficient Verifiable Computation for CKKS](https://antonioguimaraes.org/publications/batch-pack-prove.md) (ASIACRYPT 2026), goal 2: Verifiable computation for CKKS, the scheme used for machine learning over encrypted data, made more efficient: a leaner arithmetization of CKKS computations, BatchFold (a polynomial commitment over R_q with batched evaluations), packing and SparsePack. On a neural-network architecture, the proofs are several orders of magnitude smaller and faster than prior approaches. - [PETCHA: Post-quantum Efficient Transciphering with ChaCha](https://antonioguimaraes.org/publications/petcha.md) (PQCrypto 2026), goal 1: Transciphering with ChaCha instead of AES-128, so that the symmetric layer is post-quantum secure like the FHE scheme, while being up to 11.7 times faster and with over 50 times higher throughput than AES-based transciphering. - [NIST submissions: Kryptallage and Tulva](https://csrc.nist.gov/Projects/threshold-cryptography/tcall-1) (Standardization, 2026), goal 1: Two submissions to the first round of the [NIST Multi-Party Threshold Cryptography call](https://csrc.nist.gov/Projects/threshold-cryptography/tcall-1): [Kryptallage](https://csrc.nist.gov/csrc/media/Projects/threshold-cryptography/documents/TCall-1/Kryptallage-PW03.pdf), a unified multi-scheme FHE system with threshold key generation and decryption (FHE scheme switching, developed with a team of 16), and [Tulva](https://csrc.nist.gov/csrc/media/Projects/threshold-cryptography/documents/TCall-1/Tulva-PW03.pdf), the asynchronous threshold FHE above (team of 6). - [The `vfhe` library](https://vfhe.ai) (Software), goal 2: A library for zero-knowledge proofs, (verifiable) fully homomorphic encryption and related techniques, which carries this line of work into reusable code ([GitHub](https://github.com/vfhe/vfhe)).