Kidist Amde Mekonnen

Kidist Amde Mekonnen

ELLIS PhD Researcher at University of Amsterdam, IRLab

Kidist Amde Mekonnen

I am an ELLIS PhD Researcher in the Information Retrieval Lab at the University of Amsterdam, supervised by Prof. dr. Maarten de Rijke and Dr. Andrew Yates.

My research focuses on generative retrieval, recommendation, neural information retrieval, multilingual retrieval, retrieval-augmented generation, and large-scale machine learning. I study how retrieval systems can represent, generate, rank, and adapt to information needs across languages and domains. My work connects generative information retrieval with efficient learning objectives, continual memory, robust decoding, and evaluation for real-world retrieval settings.

Research Interests

  • Generative information retrieval
  • Neural information retrieval and recommender systems
  • Multilingual and low-resource retrieval
  • Continual learning for retrieval systems
  • Robust decoding and evaluation for generative retrieval
  • Retrieval-augmented generation and large-scale machine learning
  • Broader interests in generative modeling, multimodal learning, and reinforcement learning

Selected Publications

[SIGIR 2026] A Parametric Memory Head for Continual Generative Retrieval [DOI] (Full Paper, to appear)
The 49th International ACM SIGIR Conference on Research and Development in Information Retrieval
Kidist Amde Mekonnen, Yubao Tang, and Maarten de Rijke

[SIGIR 2026] Lost in Decoding? Reproducing and Stress-Testing the Look-Ahead Prior in Generative Retrieval [DOI] (Reproducibility Track, to appear)
The 49th International ACM SIGIR Conference on Research and Development in Information Retrieval
Kidist Amde Mekonnen, Yongkang Li, Yubao Tang, Simon Lupart, and Maarten de Rijke

[SIGIR 2025] Lightweight and Direct Document Relevance Optimization for Generative Information Retrieval [DOI] (Full Paper)
The 48th International ACM SIGIR Conference on Research and Development in Information Retrieval
Kidist Amde Mekonnen, Yubao Tang, and Maarten de Rijke

[ACL Findings 2025] Optimized Text Embeddings & Benchmarks for Amharic Passage Retrieval [Paper]
Findings of the Association for Computational Linguistics
Kidist Amde Mekonnen, E. Alemneh, and Maarten de Rijke

[arXiv 2024] Adv-KD: Adversarial Knowledge Distillation for Faster Diffusion Sampling [Paper]

[arXiv 2023] Balanced Face Dataset: Guiding StyleGAN for Labeled Synthetic Face Image Dataset for Underrepresented Group [Paper]

[arXiv 2024] Conditioning GAN Without Training Dataset [Paper]

Research Highlights

  • Continual generative retrieval: developing parametric-memory approaches that allow generative retrieval systems to update and retain document knowledge over time.
  • Decoding and robustness: reproducing and stress-testing look-ahead decoding priors to understand when generative retrieval gains are reliable.
  • Direct relevance optimization: designing lightweight objectives for improving document relevance in generative information retrieval.
  • Multilingual retrieval: building text embeddings and benchmarks for Amharic and other low-resource retrieval settings.
  • Recommendation and RAG: exploring generative recommendation, semantic identifiers, trie-constrained generation, and retrieval-augmented systems.

Education and Experience

  • 2023–Present, ELLIS PhD Researcher, University of Amsterdam, The Netherlands
  • Apr. 2022–Oct. 2022, Data Science Research Intern, Nanovery, Newcastle upon Tyne, United Kingdom
  • 2020–2023, MSc in Data Science, University of Trento, Italy
  • 2019–2020, MSc in Machine Intelligence, African Institute for Mathematical Sciences / AMMI, Rwanda
  • 2017–2019, Assistant Lecturer, Department of Computer Science, University of Gondar, Ethiopia
  • 2013–2017, BSc in Computer Science, University of Gondar, Ethiopia

Teaching and Service

Contact