BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//programme.europython.eu//europython-2026//speaker//7T3KC
 8
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-europython-2026-ZPCDKE-0@programme.europython.eu
DTSTART;TZID=CET:20260714T093000
DTEND;TZID=CET:20260714T110000
DESCRIPTION:Implementing high-performance deep learning models often feels 
 like a struggle between readable Python code and the low-level optimizatio
 ns required for modern GPUs and TPUs. JAX bridges this gap by treating neu
 ral networks as pure mathematical transformations. In this session\, we wi
 ll move beyond the abstractions of high-level frameworks to build a Denois
 ing Diffusion Probabilistic Model (DDPM) from the ground up.\n\nWe will ex
 plore how JAX’s functional programming paradigm is uniquely suited for t
 he stochastic nature of diffusion. You will learn how to:\n\n- Master the 
 JIT (Just-In-Time) compilation: See how @jax.jit transforms Python functio
 ns into optimized XLA kernels for massive speedups.\n- Leverage Vectorized
  Mapping: Use @jax.vmap to handle data parallelism across batches without 
 the overhead of manual loops.\n- Dissect the Diffusion Pipeline: Step thro
 ugh the forward noise process (SDEs) and the reverse denoising process (Sc
 ore-matching).\n- Manage State and PRNGs: Navigate JAX’s unique\, explic
 it handling of random number generation and stateless transformations.\n\n
 This tutorial is designed for Python developers and ML engineers who want 
 to understand the "how" and "why" behind state-of-the-art text-to-image mo
 dels. You will leave with a deep understanding of the diffusion objective 
 and the practical skills to deploy high-performance model architectures us
 ing the JAX ecosystem.
DTSTAMP:20260524T121644Z
LOCATION:Conference Hall Complex B (S4B)
SUMMARY:Let it rip a diffusion tutorial - Mai Giménez
URL:https://programme.europython.eu/europython-2026/talk/ZPCDKE/
END:VEVENT
BEGIN:VEVENT
UID:pretalx-europython-2026-ZPCDKE-1@programme.europython.eu
DTSTART;TZID=CET:20260714T111500
DTEND;TZID=CET:20260714T124500
DESCRIPTION:Implementing high-performance deep learning models often feels 
 like a struggle between readable Python code and the low-level optimizatio
 ns required for modern GPUs and TPUs. JAX bridges this gap by treating neu
 ral networks as pure mathematical transformations. In this session\, we wi
 ll move beyond the abstractions of high-level frameworks to build a Denois
 ing Diffusion Probabilistic Model (DDPM) from the ground up.\n\nWe will ex
 plore how JAX’s functional programming paradigm is uniquely suited for t
 he stochastic nature of diffusion. You will learn how to:\n\n- Master the 
 JIT (Just-In-Time) compilation: See how @jax.jit transforms Python functio
 ns into optimized XLA kernels for massive speedups.\n- Leverage Vectorized
  Mapping: Use @jax.vmap to handle data parallelism across batches without 
 the overhead of manual loops.\n- Dissect the Diffusion Pipeline: Step thro
 ugh the forward noise process (SDEs) and the reverse denoising process (Sc
 ore-matching).\n- Manage State and PRNGs: Navigate JAX’s unique\, explic
 it handling of random number generation and stateless transformations.\n\n
 This tutorial is designed for Python developers and ML engineers who want 
 to understand the "how" and "why" behind state-of-the-art text-to-image mo
 dels. You will leave with a deep understanding of the diffusion objective 
 and the practical skills to deploy high-performance model architectures us
 ing the JAX ecosystem.
DTSTAMP:20260524T121644Z
LOCATION:Conference Hall Complex B (S4B)
SUMMARY:Let it rip a diffusion tutorial - Mai Giménez
URL:https://programme.europython.eu/europython-2026/talk/ZPCDKE/
END:VEVENT
END:VCALENDAR