BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Exploring Descriptor Space for Applications in Materials Science -
  Dr Mihai-Cosmin Marinica\, Université Paris-Saclay\, CEA
DTSTART:20251022T133000Z
DTEND:20251022T143000Z
UID:TALK232585@talks.cam.ac.uk
CONTACT:Lisa Masters
DESCRIPTION:This study investigates the internal representations used in m
 achine learning potentials or deep learning models\,  focusing on the high
 -dimensional feature vectors\, also known as descriptors\, that encode ato
 mic structures. These descriptors form an inner space that appears to poss
 ess an intrinsic structure\, which we aim to exploit for the characterizat
 ion of complex energy landscapes. This structured space provides a powerfu
 l framework for studying the interactions and transformations within netwo
 rks of crystal defects\, which give rise to a remarkably diverse range of 
 defect morphologies [1].\nRather than spending effort on developing yet an
 other machine learning potential\, an already mature and widely explored t
 opic\, we will instead concentrate on inspecting and exploiting the descri
 ptor space itself. By analyzing this descriptor space\, we can: (i) identi
 fy and statistically characterize complex defect networks [1\,2]\; (ii) co
 mbine these insights with accelerated Molecular Dynamics methods\, such as
  the Bayesian Adaptive Biasing Force approach [3]\, to efficiently sample 
 intricate defect energy landscapes\; (iii) construct surrogate models that
  bypass traditional methodologies to predict demanding properties\, such a
 s vibrational entropies\, with significantly reduced computational effort 
 [4]\; and (iv) demonstrate that descriptors can encode an agnostic entropy
 \, thereby establishing a link between Gibbs and Shannon entropy and enabl
 ing a natural generalization of anharmonic free energies across a wide ran
 ge of relevant physical systems [5]. For the broad class of generalized li
 near models we show free energies can be cast as the Legendre transform of
  a high-dimensional descriptor entropy\, accurately estimated via score ma
 tching. \nThis last concept provides the first example in the literature w
 here the free energy itself can be backpropagated. We present a model agno
 stic estimator which returns meV/atom accurate\, end-to-end differentiable
  free energies over a diverse\, multi-element range of parameters. \n[1] A
 . M. Goryaeva et al. Nature Commun. 14\, 3003 (2023)\; A. M. Goryaeva et a
 l. Nature Commun. 11\, 4691 (2020)\; P. Lafourcade et al. Comp. Mater. Sci
 .  230\, 112534 (2025)\n[2] M.-C. Marinica\, A. M. Goryaeva\, T. D. Swinbu
 rne et al\, MiLaDy - Machine Learning Dynamics\, CEA Saclay\, 2015-2025:  
 https://ai-atoms.github.io/milady/ \; \n[3] A. Zhong et al\, Phys. Rev. Ma
 ter. 7\, 023802 (2023)\;  A. Zhong et al.  PRX Energy 4\, 013008 (2025)\; 
 C. Lapointe et al. Phys. Rev. Mater. 9\,  093801 (2025) \; A. Allera et al
  Nature Commun. 16\, 8367 (2025). \n[4] C. Lapointe\, et al\, Phys. Rev. M
 ater. 6\, 113803 (2022).\n[5] T. D. Swinburne et al  arXiv:2502.18191 (202
 5). \n
LOCATION:Unilever Lecture Theatre\, Yusuf Hamied Department of Chemistry
END:VEVENT
END:VCALENDAR