Amirhossein Samandar

Ph.D. candidate in Physics, Case Western Reserve University

Calibrated inference for the one universe we observe

My research centers on how we extract physical information from cosmological data, determine what the data can distinguish, and make demanding analyses computationally feasible without compromising their statistical interpretation.

Seeking postdoctoral positions starting 2027. Ph.D. expected January 2027.

Citation metrics

Citations
122
h-index
7
i10-index
6

Google Scholar, as of 7 November 2026 · View profile

One universe, one sky

We observe exactly one universe. There is no second sky to check our answers against, so knowing what we can reliably learn from the one we have is central to cosmology. This question connects my work across cosmological data analysis, statistics, probabilistic machine learning, and AI. How do we extract information, quantify uncertainty, and know when to trust an answer?

In cosmology, I investigate whether non-trivial cosmic topology leaves detectable traces in the cosmic microwave background (CMB). In statistics, I study how often a search can recover a real signal and constrain its underlying parameters. Through probabilistic machine learning, I develop methods that learn from simulations to accelerate inference while testing the reliability of their uncertainties.

I also work on calibrated Bayesian evaluation of large language models and explore how AI agents can support scientific research. Across these directions, my aim is to make scientific conclusions and the tools behind them more trustworthy, with a clear account of what the evidence supports and what remains uncertain.

Headline results

  • Detectability · first author

    The KL divergence does not, by itself, say how often a search detects a signal. I derived Gaussian likelihood-ratio distributions and calibrated full parameter searches with simulations.

  • CMB parity · first author

    Spatial topology by itself, with no parity-violating microphysics, can induce parity-odd TB and EB correlations in the CMB.

  • Tensor modes · first author

    First spin-2 eigenmodes and CMB correlation matrices for orientable compact flat spaces; tensor-mode topological information keeps growing with ℓ over the range computed.

  • LLM evaluation · second author

    A Dirichlet-posterior replacement for Pass@k gives credible intervals and more stable model rankings from far fewer samples.

Selected papers

  1. First author Lead authorPreprint

    Beyond the Kullback–Leibler divergence: the likelihood-ratio distribution as a detectability measure

    arXiv preprint · Submitted to arXiv 12 Oct 2026

    Why it matters: Lets detectability forecasts be calibrated for the model family and test statistic actually used, rather than read off the KL divergence alone.

    My role: Conceived the project and carried out the full analysis, including the analytical derivations, numerical implementation, and interpretation of the results.

  2. First author Lead author

    Cosmic topology. Part IIIb. Eigenmodes and correlation matrices of spin-2 perturbations in orientable Euclidean manifolds

    JCAP 08 (2025) 015 · 2025

    Why it matters: Suggests that future CMB polarization data could open new avenues for detecting cosmic topology.

    My role: Led the analytical derivation of spin-2 eigenmodes and CMB covariance matrices, together with the numerical implementation and analysis of the results.

  3. First author Lead author

    Cosmic topology. Part IIIa. Microwave background parity violation without parity-violating microphysics

    JCAP 11 (2024) 020 · 2024

    Why it matters: A measured TB/EB signal does not by itself imply parity-violating microphysics; spatial topology is an alternative that has to be tested.

    My role: Led the analytical derivations, numerical implementation, and analysis of the results on topology-induced CMB parity violation.

  4. Second author Lead author

    Don't Pass@k: A Bayesian Framework for Large Language Model Evaluation

    ICLR 2026, main conference · 2026

    Why it matters: Gives benchmark rankings an explicit finite-sample uncertainty, so a difference between two models can be judged against its credible interval.

    My role: As a lead author, co-developed the Bayesian framework for LLM evaluation, including posterior estimates and credible intervals for finite-sample uncertainty.

  5. Fourth author

    The topology of the Universe

    Nature Astronomy, Review Article (advance online publication) · Published online 11 Sep 2026

    Why it matters: The collaboration's reference statement on cosmic topology, in a broad-readership journal.

    My role: Contributed to the CMB analysis, prepared the relevant plots, and helped develop the discussion of covariance analysis in cosmic-topology studies.

All publications (13 papers: 7 published, 1 accepted, 5 preprints; 1 more in preparation, not counted)

Research themes

Four areas, one question: how much can a single, noisy measurement really tell us?

  1. Cosmology

    Could space be finite? I compute how its topology would imprint the CMB and build pipelines to search Planck data for it.

  2. Statistics and data analysis

    Detection probabilities, not just averages: likelihood-ratio tests calibrated for the search actually performed.

  3. Probabilistic machine learning

    Neural emulators and amortized inference that make demanding analyses fast, checked against exact likelihoods (in development).

  4. LLMs and agentic AI

    Bayesian evaluation of language models, inference-time scaling, and AI agents in research, with checks on their reasoning and code.

Software

  • CMBtopology

    GPU-parallelized package computing CMB covariance matrices for compact Euclidean topologies; used across COMPACT for tensor-mode likelihood analysis.

    My role: Developed the GPU-parallelized Python/HPC implementation.

    Python · CUDA · HPC · v0.1.0 (2025)

  • Scorio

    Open-source Bayesian evaluation toolkit for LLMs: posterior estimates and credible intervals in place of Pass@k.

    My role: Contributed to the development of the new version.

    Python · Julia · v0.2.2

News

  1. Google Scholar citations reached 122, with an h-index of 7. Google Scholar

  2. Posted first-author preprint “Beyond the Kullback–Leibler divergence: the likelihood-ratio distribution as a detectability measure.” arXiv:2610.17186

  3. COMPACT preprint Cosmic topology Part IId (lens-space eigenmodes) posted. arXiv:2610.12418

  4. COMPACT review “The topology of the Universe” published online in Nature Astronomy. DOI

  5. Cosmic topology Part IIb published in JCAP 07 (2026) 004. DOI

All news

Collaborations and service

  • COMPACT Collaboration member (cosmic topology).
  • LiteBIRD adjoint member: topology forecasts using E-mode polarization with temperature.
  • Mentoring undergraduate researchers one-on-one.
  • Art Walker Award for Outstanding Service, CWRU Physics (2026). References available on request.