Researcher · University of Cyprus

Yobsan Bayisa Leta.

PhD Student / MSCA Doctoral Researcher

BASED INNicosia, Cyprus

Making vision models more reliable in the world beyond the benchmark.

I am a PhD student at the University of Cyprus, conducting research at the KIOS Research and Innovation Center of Excellence. As part of the TIRAMISU MSCA Doctoral Network, I study robust and efficient neural networks for edge AI.

My research connects computer vision, robust deep learning, and neural architecture search. I am particularly interested in semantic segmentation under adverse conditions, and in models that balance accuracy with the memory, energy, and latency constraints of edge devices.

Robust Edge AIComputer VisionNeural Architecture Search

Explore my research Curriculum vitae

Latest news01

All updates

2026
LATESTLiDSeg at BMVC 2026
LiDSeg, our work on efficient semantic segmentation in adverse weather, is listed as in press at the British Machine Vision Conference.
2025-06-08
Mobility-aware edge computing
Our HAC-LSTM paper on service placement and task offloading appears at IEEE ICC 2025.
2025-05-15
A new chapter at KIOS
I joined the University of Cyprus and the TIRAMISU MSCA Doctoral Network to research robust neural networks for edge AI.

Latest publications02

All publications

BMVC 2026

LiDSeg: Linear DehazeFormer Guidance for Efficient Semantic Segmentation in Adverse Weather Conditions

Yobsan Bayisa Leta, Christos Kyrkou, Maria K. Michael, Theocharis Theocharides

British Machine Vision Conference (BMVC) · In press · 2026

Paper record Code

Research summary

LiDSeg adds a lightweight guidance encoder to semantic segmentation models for adverse weather. Auxiliary image reconstruction supports training; at inference, the reconstruction decoder is removed while the guidance encoder and gated fusion remain. The framework supports multiple segmentation backbones and evaluation across fog, rain, snow, and night conditions.

IEEE ICC 2025

HAC-LSTM: Mobility-Aware Joint Service Placement and Task Offloading Scheme in MEC

Surafel Kifetew Woldeyes, Yongmin Zhang, Wei Wang, Leta Yobsan Bayisa

ICC 2025 - IEEE International Conference on Communications · 2025

DOI

Research summary

A mobility-aware approach to joint service placement and task offloading in mobile edge computing. This work studies how service decisions can account for changing mobility patterns.

Journal of Big Data

Online variational Gaussian process for time series data

Weidong Wang, Mian Muhammad Yasir Khalil, Leta Yobsan Bayisa

Journal of Big Data · 2024

DOI

Research summary

Research on online variational Gaussian processes for time-series data, extending probabilistic modeling to observations that arrive over time.

Research interests03

Robust visual perception

Semantic segmentation that can handle adverse weather, distribution shifts, and challenging environments.

Intelligence at the edge

Lightweight neural networks that balance accuracy with practical limits on memory, energy, and latency.

Learning & architecture

Multitask representations and architecture search for efficient, adaptable visual models.

Let’s connect.

For research conversations, questions, and collaboration.

Get in touch