BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Palaner//palaner Single Event//EN
METHOD:PUBLISH
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:meetup-80e0bf8c63@palaner.app
DTSTAMP:20260722T112941Z
DTSTART:20260727T163000Z
DTEND:20260727T180000Z
SUMMARY:July 27 - London AI\, ML\, and Computer Vision Meetup
LOCATION:Skempton Building\, Imperial College
DESCRIPTION:Saved in Palaner as: saved\n\nPalaner match: 32% · why: lots of
  people going\n\nJoin our in-person meetup to hear talks from experts on c
 utting-edge topics across AI\, ML\, and computer vision.\n\nRegister to re
 serve your seat!\n\nDate\, Time and Location\n\nJul 27\, 2026\n5:30 PM - 8
 :30 PM BST\nImperial College London\, Skempton Building (LT201)\, South Ke
 nsington\, London SW7 2AZ\n\nUniLight: Unified Multi-Modal Lighting Repres
 entation\n\nLighting has a strong influence on visual appearance\, yet und
 erstanding and representing lighting in images remains notoriously difficu
 lt. UniLight introduces a joint latent space to unify previously incompati
 ble lighting representation - environment maps\, images\, irradiance and t
 ext descriptions.\n\nModality-specific encoders are trained contrastively 
 to align their representations\, with an auxiliary spherical-harmonics pre
 diction task reinforcing directional understanding. Our joint lighting emb
 edding enables applications such as retrieval\, example-based light contro
 l during image generation\, and environment map generation from various mo
 dalities.\n\nAbout the Speaker\n\nZitian Zhang - is a PhD candidate in Com
 puter Science at Université Laval\, and a research scientist intern in Ado
 be Research London. He focuses on image understanding\, generation\, and l
 ighting representations through foundation models.\n\nLoST: Level of Seman
 tics Tokenization for 3D Shapes\n\nTokenization is fundamental to generati
 ve modeling and especially important for autoregressive 3D generation. How
 ever\, current 3D shape tokenizers rely on geometric level-of-detail hiera
 rchies that are token-inefficient and poorly aligned with semantic structu
 re.\n\nWe propose Level-of-Semantics Tokenization (LoST)\, which orders to
 kens by semantic salience so early tokens produce complete\, plausible sha
 pes and later tokens refine detailed geometry and semantics. LoST is train
 ed with Relational Inter-Distance Alignment (RIDA)\, a semantic alignment 
 loss that matches relationships in 3D shape latent space to those in DINO 
 feature space.\n\nExperiments show that LoST achieves state-of-the-art rec
 onstruction and efficient high-quality AR 3D generation while using only 0
 .1%–10% of the tokens required by prior methods.\n\nAbout the Speaker\n\nN
 iladri Dutt - is an ELLIS PhD student at University College London (UCL)\,
  sponsored by Adobe Research. He is advised by Prof Niloy Mitra (UCL) and 
 Duygu Ceylan (Adobe).\n\nMaterial selection in 2D and beyond - methods\, t
 ricks and applications\n\nIn this talk\, we'll explore reasoning about ima
 ges from a material-centric perspective\, namely through the lens of mater
 ial understanding. Materials distinguish themselves by their response to l
 ight\, which is governed and modelled through physical properties like rou
 ghness or gloss - however\, understanding such properties is a non-trivial
  task for current algorithms and models.\n\nWe'll see how we can select ma
 terials similar to a given query material\, significantly improve selectio
 n fidelity and eventually even venture beyond 2D\, to enable selection in 
 the 3D domain.\n\nAbout the Speaker\n\nMichael Fischer - is a research sci
 entist at Adobe research London. He obtained his PhD from University Colle
 ge London (UCL)\, advised by Niloy Mitra and Tobias Ritschel. Michael has 
 authored several top-tier publications (CVPR\, ICCV\, SIGGRAPH\, ...) and 
 is a recipient of both the Meta PhD scholarship and the Rabin Ezra scholar
 ship as well as the Eurographics PhD Thesis award 2026.\n\nLessons from th
 e Trenches of Agentic Engineering\n\nA candid lessons-learned from running
  an agentic engineering consultancy with clients ranging from federal gove
 rnments to early-stage AI startups. I'll cover what's held up under real p
 roduction pressure\, what I tried and abandoned\, and the approaches that 
 are quietly dead but still being sold. Expect specifics\, opinions\, and a
  few uncomfortable conclusions.\n\nAbout the Speaker\n\nJohn Adeojo - runs
  Brainqub3 an agentic engineering consultancy serving clients from federal
  governments to early-stage AI startups\, and recently served briefly as C
 TO of a pre-seed AI startup.
URL:https://www.meetup.com/london-ai-machine-learning-data-science/events/3
 15091215/
CATEGORIES:palaner,saved
STATUS:TENTATIVE
END:VEVENT
END:VCALENDAR