Meta tags:
author= Ignacio Peis;
description= A simple, whitespace theme for academics. Based on [*folio](htt????/github.com/bogoli/-folio) design. ;
keywords= artificial intellingence, research, deep generative models;
Headings (most frequently used words):
ignacio, peis, news, selected, publications,
Text of the page (most frequently used words):
and (43), the (22), learning (13), models (12), with (11), for (11), data (10), based (10), #ignacio (9), peis (9), latent (9), 2026 (8), missing (8), our (8), machine (7), generative (7), paper (7), research (6), true (6), that (6), inference (6), proceedings (6), diffusion (6), from (6), this (5), hyper (5), poster (5), jul (5), conference (5), pablo (5), over (5), functions (5), vamoh (5), more (4), selected (4), neural (4), processing (4), systems (4), imputation (4), deep (4), variational (4), been (4), existing (4), international (4), koyuncu (4), batuhan (4), valera (4), isabel (4), representations (4), hypernetworks (4), transforming (4), inr (4), university (4), about (3), bibtex_show (3), year (3), information (3), booktitle (3), lobato (3), miguel (3), author (3), hierarchical (3), hamiltonian (3), monte (3), carlo (3), title (3), inproceedings (3), are (3), gaussian (3), present (3), vaem (3), experiments (3), show (3), tasks (3), approach (3), slides (3), code (3), pdf (3), bib (3), abs (3), olmos (3), mixture (3), distributions (3), approaches (3), inrs (3), such (3), work (3), can (3), icml (3), 2025 (3), frellsen (3), jes (3), framework (3), transformer (3), will (3), postdoctoral (3), two (3), where (3), engineering (3), group (3), denmark (3), dtu (3), pages (2), you (2), 2022 (2), advances (2), chao (2), acquisition (2), autoencoders (2), vaes (2), have (2), however (2), within (2), vae (2), methods (2), using (2), only (2), variables (2), approximations (2), limitations (2), model (2), improved (2), supervised (2), features (2), sampling (2), when (2), video (2), josé (2), hernández (2), 2023 (2), 40th (2), sanchez (2), martin (2), hypergenerators (2), implicit (2), propose (2), function (2), computationally (2), novel (2), capabilities (2), continuous (2), flow (2), prior (2), expressive (2), range (2), generation (2), better (2), while (2), 42nd (2), scalability (2), decoder (2), trained (2), via (2), publications (2), has (2), accepted (2), modelling (2), info (2), soon (2), mathematics (2), seoul (2), probml (2), probabilistic (2), was (2), telecommunications (2), signal (2), uc3m (2), prof (2), department (2), current (2), applied (2), researcher (2), technical (2), section (2), cognitive (2), copyright, powered, theme, hosted, last, updated, july, github, folio, jekyll, not, hesitate, writing, email, want, know, hern, ndez, jos, peis2022missing, recently, highly, successful, imputing, acquiring, heterogeneous, specific, application, domain, restricted, one, layer, strictly, posterior, address, these, mixed, type, incomplete, uses, automatic, parameter, tuning, approximate, outperforms, baselines, finally, also, efficiently, computing, gain, acquired, superior, alternatives, neurips, koyuncu2023variational, recent, build, spaces, they, intensive, dealing, directly, cannot, tackle, them, named, combines, modeling, addition, relies, normalizing, define, parametrize, log, likelihood, gives, high, capability, interpretability, through, diverse, types, images, voxels, climate, effectively, learn, rich, furthermore, perform, related, conditional, super, resolution, painting, well, than, previous, being, less, demanding, https, mlr, press, v267, peis25a, html, url, pmlr, publisher, month, series, 267, volume, singh, aarti, fazel, maryam, hsu, daniel, lacoste, julien, simon, berkenkamp, felix, maharaj, tegan, wagstaff, kiri, zhu, jerry, editor, 48714, 48733, peis2025hyper, introduce, integrating, into, variable, unlike, rely, mlp, method, employs, generate, parameters, addressing, both, representation, capacity, computational, efficiency, extends, ldms, replacing, standard, decoders, hypernetwork, which, either, scratch, strategy, fine, tunes, freezing, pre, space, enables, efficient, adaptation, without, requiring, full, retraining, validate, across, multiple, modalities, demonstrating, expressiveness, generalization, findings, establish, unified, flexible, structured, tomorrow, etsii, universidad, sevilla, give, oral, session, part, andaluzia, dec, tmlr, scalable, physical, source, field, jan, awarded, lead, line, meta, granada, institute, caixa, foundation, junior, leader, fellowship, mar, papers, see, south, korea, may, find, posters, denoising, uncertainty, distributional, news, completed, phd, previously, intern, worked, obtained, msc, degrees, bsc, degree, ugr, cambridge, antonio, artés, rodríguez, short, bio, interested, connection, between, lies, creating, increasing, their, robustness, developing, several, fields, like, neuroscience, physics, psychiatry, embark, fascinating, projects, pioneer, centre, artificial, intelligence, computer, science, ipeaz, 2800, kgs, lyngby, richard, petersens, plads, building, 321, office, 220, contact, ctrl, repositories, toggle, navigation,
Text of the page (random words):
ignacio peis toggle navigation about current publications repositories cv ctrl k ignacio peis postdoctoral researcher contact building 321 office 220 richard petersens plads 2800 kgs lyngby denmark ipeaz dtu dk section for cognitive systems department of applied mathematics and computer science technical university of denmark dtu pioneer centre for artificial intelligence about i am a postdoctoral researcher at the section for cognitive systems in technical university of denmark dtu i work within jes frellsen s group where we embark on a range of fascinating projects research i am interested in the connection between deep learning and probabilistic modelling my current research lies in creating more expressive generative models increasing their robustness and developing better inference methods my work has been applied to several fields like neuroscience physics or psychiatry short bio i completed my phd in probabilistic machine learning at uc3m where i was supervised by prof antonio artés rodríguez and dr pablo m olmos from the signal processing group previously i was a research intern at the machine learning group in the department of the engineering university of cambridge where i worked with prof josé miguel hernández lobato i obtained two msc degrees in telecommunications engineering and signal processing from uc3m and a bsc degree in telecommunications engineering from ugr news jul 05 2026 find me in seoul at probml 2026 at 14 00 i will present two posters denoising uncertainty via gaussian distributional diffusion models 1 and latent diffusion for missing data 2 paper 1 poster 1 paper 2 poster 2 may 08 2026 two papers accepted at probml 2026 see you in seoul south korea more info soon paper mar 10 2026 i have been awarded a junior leader postdoctoral fellowship from the la caixa foundation i will lead a research line on flow based generative modelling and meta learning at the institute of mathematics university of granada more info soon jan 15 2026 our paper scalable physical source to field inference with hypernetworks has been accepted in tmlr paper dec 18 2025 tomorrow at the etsii universidad de sevilla i will give an oral session on our icml paper hyper transforming latent diffusion models as part of the andaluzia conference paper slides selected publications icml hyper transforming latent diffusion models ignacio peis batuhan koyuncu isabel valera and jes frellsen in proceedings of the 42nd international conference on machine learning 13 19 jul 2025 abs bib pdf code poster slides we introduce a novel generative framework for functions by integrating implicit neural representations inrs and transformer based hypernetworks into latent variable models unlike prior approaches that rely on mlp based hypernetworks with scalability limitations our method employs a transformer based decoder to generate inr parameters from latent variables addressing both representation capacity and computational efficiency our framework extends latent diffusion models ldms to inr generation by replacing standard decoders with a transformer based hypernetwork which can be trained either from scratch or via hyper transforming a strategy that fine tunes only the decoder while freezing the pre trained latent space this enables efficient adaptation of existing generative models to inr based representations without requiring full retraining we validate our approach across multiple modalities demonstrating improved scalability expressiveness and generalization over existing inr based generative models our findings establish a unified and flexible framework for learning structured function representations inproceedings peis2025hyper title hyper transforming latent diffusion models author peis ignacio and koyuncu batuhan and valera isabel and frellsen jes booktitle proceedings of the 42nd international conference on machine learning pages 48714 48733 year 2025 editor singh aarti and fazel maryam and hsu daniel and lacoste julien simon and berkenkamp felix and maharaj tegan and wagstaff kiri and zhu jerry volume 267 series proceedings of machine learning research month 13 19 jul publisher pmlr url https proceedings mlr press v267 peis25a html bibtex_show true selected true icml variational mixture of hypergenerators for learning distributions over functions batuhan koyuncu pablo sanchez martin ignacio peis pablo m olmos and isabel valera in proceedings of the 40th international conference on machine learning 13 19 jul 2023 abs bib pdf video code poster recent approaches build on implicit neural representations inrs to propose generative models over function spaces however they are computationally intensive when dealing with inference tasks such as missing data imputation or directly cannot tackle them in this work we propose a novel deep generative model named vamoh vamoh combines the capabilities of modeling continuous functions using inrs and the inference capabilities of variational autoencoders vaes in addition vamoh relies on a normalizing flow to define the prior and a mixture of hypernetworks to parametrize the data log likelihood this gives vamoh a high expressive capability and interpretability through experiments on a diverse range of data types such as images voxels and climate data we show that vamoh can effectively learn rich distributions over continuous functions furthermore it can perform inference related tasks such as conditional super resolution generation and in painting as well or better than previous approaches while being less computationally demanding inproceedings koyuncu2023variational title variational mixture of hypergenerators for learning distributions over functions author koyuncu batuhan and sanchez martin pablo and peis ignacio and olmos pablo m and valera isabel booktitle proceedings of the 40th b international conference on machine learning b year 2023 selected true bibtex_show true neurips missing data imputation and acquisition with deep hierarchical models and hamiltonian monte carlo ignacio peis chao ma and josé miguel hernández lobato in advances in neural information processing systems 35 13 19 jul 2022 abs bib pdf video code poster slides variational autoencoders vaes have recently been highly successful at imputing and acquiring heterogeneous missing data however within this specific application domain existing vae methods are restricted by using only one layer of latent variables and strictly gaussian posterior approximations to address these limitations we present hh vaem a hierarchical vae model for mixed type incomplete data that uses hamiltonian monte carlo with automatic hyper parameter tuning for improved approximate inference our experiments show that hh vaem outperforms existing baselines in the tasks of missing data imputation and supervised learning with missing features finally we also present a sampling based approach for efficiently computing the information gain when missing features are to be acquired with hh vaem our experiments show that this sampling based approach is superior to alternatives based on gaussian approximations inproceedings peis2022missing title missing data imputation and acquisition with deep hierarchical models and hamiltonian monte carlo author peis ignacio and ma chao and hern a ndez lobato jos e miguel booktitle advances in b neural information processing systems b 35 year 2022 selected true bibtex_show true do not hesitate in writing an email if you want to know more about my research copyright 2026 ignacio peis powered by jekyll with al folio theme hosted by github pages last updated july 03 2026
|