Marcin Abram

Publications

Ten peer-reviewed articles, six working papers, a book chapter, and a patent. Citation counts are on Google Scholar.

Machine learning

2020 to 2026

  • Emergent inference-time semantic contamination via in-context priming M. Abram arXiv:2604.04043 (2026) Five culturally loaded numbers given as few-shot examples shift a capable model's answers to an unrelated prompt toward darker themes. A smaller model does not shift.
  • Toward evaluation frameworks for multi-agent scientific AI systems M. Abram arXiv:2603.26718 (2026) How to benchmark scientific agents when retrieval can pass for reasoning, training data is contaminated, and novel problems have no ground truth. Includes a dataset of new research ideas and interviews with researchers in quantum science.
  • Why does in-context learning fail sometimes? Evaluating in-context learning on open and closed questions X. Li, H. Tang, S. Chen, Z. Wang, R. Chen, M. Abram arXiv:2407.02028 (2024) A context closely aligned with the topic does not always help more than a loosely related one, especially for open, difficult, or novel questions.
  • Context matters: data-efficient augmentation of large language models for scientific applications X. Li, H. Tang, S. Chen, Z. Wang, A. Maravi, M. Abram arXiv:2312.07069 (2023) The relation between the relevance of a context and the quality of the answer is not linear.
  • Emulating quantum dynamics with neural networks via knowledge distillation Y. Yao, C. Cao, S. Haas, M. Agarwal, D. Khanna, M. Abram Frontiers in Materials 9, 1060744 (2023) An emulator trained on a curriculum of simple examples generalizes to wave packets and potentials it has not seen.
  • Inferring topological transitions in pattern-forming processes with self-supervised learning M. Abram, K. Burghardt, G. Ver Steeg, A. Galstyan, R. Dingreville npj Computational Materials 8, 205 (2022) The uncertainty of a network trained on the inverse problem marks where a material's microstructure changes regime, with no labels.
  • Performance weighting for robust federated learning against corrupted sources D. Stripelis, M. Abram, J. L. Ambite arXiv:2205.01184 (2022) Weighting each participant by measured performance keeps a federation useful when some participants train on corrupted labels.
  • Model-free real-time autonomous energy management for a residential multi-carrier energy system: a deep reinforcement learning approach Y. Ye, D. Qiu, J. Ward, M. Abram Proceedings of IJCAI 2020, pp. 339–346

Decentralized systems

2019 to 2026

  • Distributed computer system and method of operation thereof J. Ward, M. Abram, D. Honerkamp US Patent 12,749,011. Filed in June 2021 with a priority date of June 2020, granted on 29 September 2026. The application was published as US 2023/0316127 A1 (2023), EP 3929742 A1, and WO 2021/260517 A1 (2021) Collective learning without a central coordinator. Worker nodes are grouped into clusters, each cluster trains a model on its own ledger without sharing raw data, and a main ledger coordinates the whole.
  • Democratising blockchain: a minimal agency consensus model M. Abram, D. Galindo, D. Honerkamp, J. Ward, J.-M. Wong Presented at Tokenomics 2019, Paris. arXiv:2006.05390 (2020) A consensus protocol that combines a directed acyclic graph with a chain of blocks, so that blocks are built collectively and no round leader can reorder or withhold transactions.

Condensed matter physics

2013 to 2024

  • Hofstadter butterflies in topological insulators L. Li, M. Abram, A. Prem, S. Haas Chapter in Recent Topics on Topology: From Classical to Modern Applications, IntechOpen (2024)
  • Antiferromagnetism, charge density wave, and d-wave superconductivity in the extended t–J–U model: role of intersite Coulomb interaction and a critical overview of renormalized mean field theory M. Abram, M. Zegrodnik, J. Spałek Journal of Physics: Condensed Matter 29, 365602 (2017)
  • Tricritical wings in UGe2: a microscopic interpretation M. Abram, M. M. Wysokiński, J. Spałek Journal of Magnetism and Magnetic Materials 400, 27 (2016)
  • Criticalities in the itinerant ferromagnet UGe2 M. M. Wysokiński, M. Abram, J. Spałek Physical Review B 91, 081108(R) (2015)
  • Ferromagnetism in UGe2: a microscopic model M. M. Wysokiński, M. Abram, J. Spałek Physical Review B 90, 081114(R) (2014)
  • t–t′–J–U model in mean-field approximation: coexistence of superconductivity and antiferromagnetism M. Abram Acta Physica Polonica A 126, A-25 (2014)
  • Model of hard spheroplatelets near a hard wall A. Kapanowski, M. Abram Physical Review E 89, 062503 (2014)
  • d-wave superconductivity and its coexistence with antiferromagnetism in the t–J–U model: statistically consistent Gutzwiller approach M. Abram, J. Kaczmarczyk, J. Jędrak, J. Spałek Physical Review B 88, 094502 (2013)

Reviewing

I review for Machine Learning: Science and Technology, New Journal of Physics, Journal of Physics: Condensed Matter, Journal of Physics: Materials, Superconductor Science and Technology, Physica Scripta, and IOP SciNotes. In 2020, the Journal of Physics: Condensed Matter named me an Outstanding Reviewer, an award given to about one percent of its reviewers.