Institution profile

University of Malta

Academic institutioneurope · mt
Official website
Research library40linked papers
Opportunities0open roles
Selected work

Representative Papers

Enhancing Object Detection with Privileged Information: A Model-Agnostic Teacher-Student Approach

Jan 05, 2026arXiv.org

This work investigates how to leverage fine-grained privileged information—such as masks, saliency maps, and depth cues—available only during training to enhance object detection performance without increasing inference complexity. To this end, it systematically introduces the Learning Using Privileged Information (LUPI) paradigm into object detection for the first time, proposing a model-agnostic teacher–student distillation framework in which a teacher network fuses multimodal privileged signals to guide the training of a student detector. The approach requires no architectural modifications at inference time and integrates seamlessly with mainstream detectors. Experiments on benchmarks including Pascal VOC 2012 and UAV-based litter detection demonstrate consistent and significant improvements in detection accuracy—particularly for medium and large objects—without any increase in model size or inference overhead, thereby validating its generality and effectiveness.

1 citationsRead paper

Programmable Cellular Automata

Sep 05, 2026

研究通过将元胞自动机表示为Python代码并模块化,解决了创建有效局部规则困难的问题,并探索了全局函数的作用以减少解决问题的迭代次数。

0 citationsRead paper

The Time Value of Evolution

Aug 13, 2026

This work addresses the limitation in evolutionary search where optimizing immediate rewards often overlooks the long-term adaptive potential of weak offspring, thereby constraining search efficiency. The authors model evolution as a finite-horizon Markov decision process and introduce the Lineage Value Policy Gradient (LVPG) framework. LVPG employs a shared generative backbone with decoupled policy heads to dynamically adjust mutation strength and formalizes, for the first time, the concept of “temporal value of evolution.” It incorporates a path-based long-term credit assignment mechanism to evaluate the long-term promise of mutational lineages. Experiments demonstrate that under identical resource constraints, LVPG achieves a 0.394 Sharpe-unit improvement in best validation-set AUC, exhibits fewer transient degradations, recovers more robustly, and significantly accelerates the discovery of high-performing strategies within limited computational budgets.

0 citationsRead paper
Recent publications

Latest Papers

Programmable Cellular Automata

Sep 05, 2026

研究通过将元胞自动机表示为Python代码并模块化,解决了创建有效局部规则困难的问题,并探索了全局函数的作用以减少解决问题的迭代次数。

0 citationsRead paper

The Time Value of Evolution

Aug 13, 2026

This work addresses the limitation in evolutionary search where optimizing immediate rewards often overlooks the long-term adaptive potential of weak offspring, thereby constraining search efficiency. The authors model evolution as a finite-horizon Markov decision process and introduce the Lineage Value Policy Gradient (LVPG) framework. LVPG employs a shared generative backbone with decoupled policy heads to dynamically adjust mutation strength and formalizes, for the first time, the concept of “temporal value of evolution.” It incorporates a path-based long-term credit assignment mechanism to evaluate the long-term promise of mutational lineages. Experiments demonstrate that under identical resource constraints, LVPG achieves a 0.394 Sharpe-unit improvement in best validation-set AUC, exhibits fewer transient degradations, recovers more robustly, and significantly accelerates the discovery of high-performing strategies within limited computational budgets.

0 citationsRead paper

ELMER: Evolutionary Language Model that Explores and Refines

Aug 10, 2026

This work addresses the limitations of existing program evolution methods, which struggle to effectively control the magnitude of behavioral variation and rely on syntactic edit distances that poorly capture semantic change. To overcome these challenges, the authors propose a large language model–based program evolution framework that introduces natural language as a guided, executable intermediate representation. By fine-tuning Qwen3-8B with task-conditioned supervision and orthogonal Direct Preference Optimization (oDPO), the framework enables conditional semantic variation and bidirectional compilation between natural language and a domain-specific language (GPTL). Evaluated under a fixed budget of 252 search iterations, the approach significantly improves behavioral alignment and search efficiency, with natural language representations achieving optimal fitness retention and enabling synergistic optimization of behavioral displacement and semantic fidelity.

0 citationsRead paper