Get Started
Home
Topics
Search
Library
Research questionHow can we infer Trans-Neptunian Object compositions and grain sizes from spectra despite multimodal uncertainty and real-data bias?Reflectance spectra can correspond to multiple compositions and grain sizes, making the inverse problem intrinsically ambiguous. Models trained on synthetic spectra may also become blind or biased when applied to real observations.
AI
Machine Learning
Research Paper
Statistical Machine Learning
Latest papersRecent research connected to this question, newest first.TNFlow: Amortized Posterior Inference for Trans-Neptunian Object Surface CompositionThe source evaluates a transformer and normalizing-flow inverse model trained on synthetic spectra from the Shkuratov radiative transfer model. Evidence includes synthetic test results and qualitative tests on JWST spectra; the source does not resolve whether real-data failures arise from simulator fidelity or training-set coverage.research paper · Sep 3, 2026
Related questions
How can we infer specimen thickness and crystal mistilt from PACBED images with few simulator evaluations?How can spectral compressive imaging models adapt to unseen optical configurations without heavy computation?How can diffusion priors reconstruct images from linear measurements while balancing fidelity and diversity?How can we estimate simulator-parameter distributions that reproduce observations without inaccurate fixed likelihood surrogates?