Lyes Saad Saoud

Research profile

Research summary

Lyes Saad Saoud is a researcher at Khalifa University of Science and Technology (United Arab Emirates) whose reviewed work sits at the intersection of artificial life, evolutionary robotics, and machine learning. The Evolve catalog holds three 2026 first-author arXiv papers, each with him as corresponding author. 'The Cost of Becoming' benchmarks developmental encoding, showing it increases phenotypic diversity while reducing genotype-phenotype locality, viability, and recombination robustness. 'Machine Zygote' uses a simulated diallel cross and germline swaps to demonstrate causal biparental heredity before learning in a germline-soma artificial agent, under a preregistered design. 'Settling' formalizes conditional-mean collapse on non-convex validity sets and proposes a deterministic equilibrium-refinement operator, though its evidence remains a synthetic 2D diagnostic. The OpenAlex-linked snapshot reports an h-index of 16 and an i10-index of 19, with 37 works in the last three years and yearly counts climbing from single digits to 14 in each of 2025 and 2026. Topical tags span neural networks and applications (14), advanced neural network applications (10), water quality monitoring technologies (10), solar radiation and photovoltaics (9), and smart agriculture and AI (8), indicating applied modeling alongside core theory. Reproducibility averages 70 across the three assessed papers, and the record overall depicts a strongly solo, high-tempo, method-driven profile.

Research keywords

Highlights

• Developmental encoding: diversity gains at the cost of locality

  A first-authored, corresponding-authored benchmark study finds developmental encoding increases phenotypic diversity in evolved robots but reduces genotype-phenotype locality, viability, and recombination robustness. Evolve rates it solid but narrowly scoped, with high reproducibility readiness.

• Machine Zygote: causal biparental heredity in artificial agents

  Using a simulated diallel cross and germline swaps, this preregistered study demonstrates causal biparental heredity before learning in a germline-soma artificial agent; Evolve notes the rigorous design, while the observed effects are expected under the deterministic setup.

• Settling: equilibrium inference for non-convex validity sets

  The paper formalizes conditional-mean collapse and proposes a deterministic equilibrium-refinement operator for non-convex validity sets; Evolve credits its honest scoping but notes the evidence rests on a synthetic 2D diagnostic, with a stylized stochastic baseline winning on success.

Timeline

1. 2010Earliest works in the snapshot: 2 publications recorded in the bibliometric record.
2. 2023Output accelerates to 11 works, the largest single-year count recorded to that point.
3. 20249 works recorded; the high-tempo phase established in 2023 continues.
4. 202514 works, matching the record's annual peak.
5. 202614 works again, including three first-author arXiv papers reviewed by Evolve on developmental encoding, artificial heredity, and equilibrium inference.

Papers in the Evolve catalog

• 2026-09-17first authorcorresponding

  The Cost of Becoming: Developmental Encoding Increases Phenotypic Diversity but Reduces Locality and Recombination Robustness in Evolved Robots

• 2026-09-15first authorcorresponding

  Machine Zygote: Causal Biparental Heredity Before Learning in a Germline--Soma Artificial Agent

• 2026-09-09first authorcorresponding

  Settling: Equilibrium Inference for Non-Convex Validity Sets

Frequent co-authors