Vincent Adam
@vincentadam87
Followers
339
Following
152
Media
6
Statuses
93
Machine Learning Research at @UPFBarcelona
Barcelona
Joined June 2012
Our #NeurIPS2021 paper "Dual Parameterization of Sparse Variational Gaussian Processes" is on Arxiv. We introduce an alternative 'dual' parameterization for SVGP leading to faster inference and learning. with @edchangy, @arnosolin & @EmtiyazKhan. (1/n)
2
13
97
RT @RLSummerSchool: And that's a wrap for #RLSS2023! After the chess tournament on Tuesday evening, yesterday we had our last lectures on n….
0
5
0
Generate videos in just a few seconds. Try Grok Imagine, free for a limited time.
358
634
2K
RT @RLSummerSchool: 🎉Exciting news! Applications for the Reinforcement Learning Summer School Barcelona 2023 are now open! Don't miss your….
0
14
0
We are hosting a. *Reinforcement Learning Summer School* (RLSS) 🤖.at University Pompeu Fabra in 🌴Barcelona 🌞this year . Great place, great topics, great line-up. Spread the word!.
We are thrilled to announce that the Reinforcement Learning Summer School 2023 will take place in Barcelona from June 26th to July 5th. It will provide an in-depth introduction to RL principles and algorithms. Applications will open soon. Stay tuned!.
0
2
7
Sharing notebooks I made:.Topic: gentle introduction to deep learning with jax. Requires: little to no coding experience with Python. Content: Defining & optimizing losses via (stochastic) gradient descent, from linear regression to image classification.
github.com
Introductory notebooks to deep learning using jax. Contribute to vincentadam87/intro_to_jax development by creating an account on GitHub.
0
1
11
RT @VinFL: New great day at the Reinforcement Learning summer school with @herkevanhoof , @thomaskipf , @pcastr and a practical session led….
0
6
0
Looking forward. I'll talk about optimization methods for Gaussian variational inference for Gaussian process models (and sparse extensions).
The next speaker in the virtual seminar series is Vincent Adam, who will give a talk on "Dual Parameterization of Sparse Variational Gaussian Processes" - Zoom links have been sent, but you can still find us on YouTube at the scheduled time!.
0
1
10
[Life update] I have just moved to Barcelona 😎 and joined the AI/ML group at @DTIC_UPF. Looking forward to 2 years of serious RL theory with @neu_rips @vicen__gomez, Anders Jonsson and the team. I'll be an independent post-doctoral researcher on a Spanish grant (Maria Zambrano).
5
1
64
RT @arnosolin: We are presenting 'Dual Parameterization of Sparse Variational Gaussian Processes' \w @vincentadam87, @edchangy, @EmtiyazKha….
0
11
0
For more details, check our paper and implementation based on GPflow!. Code : Arxiv : (5/n).
arxiv.org
Sparse variational Gaussian process (SVGP) methods are a common choice for non-conjugate Gaussian process inference because of their computational benefits. In this paper, we improve their...
1
0
7
The dual parameterization leads to faster inference:.natural gradients updates take a very simple form and require only local gradients of the expected log-likelihood terms in the variational objective (ELBO) - no need to differentiate the KL term. (4/n)
1
0
2
The dual parameterization allows to derive a better objective (tighter bound) in a Variational Expectation-Maximization (VEM) algorithm for learning the kernel parameters of GP models. This tighter bound leads to faster learning. We call our extended VEM algorithm t-SVGP. (3/n)
1
0
1
The 'dual' parameterization of the posterior process in SVGP shares the site-based structure of the approximation used in Expectation-Propagation, separating prior and data contributions, with 2n parameters. Sites can be tied to reduce storage complexity to O(m^2). (2/n)
1
0
2
I m happy to announce I will never have a best reviewer award. I always ask for at most 3 papers. Conferences sum the good AC reports. I will never make the cut.
1
0
7
RT @SecondmindLabs: @EmtiyazKhan recently gave a great talk at Secondmind Labs, and he was kind enough to allow us to share it with everyon….
0
3
0
RT @SecondmindLabs: Secondmind is opening a Senior Research Engineer position to work on our Gaussian process and Bayesian optimisation ope….
0
9
0
In brief it is a (opinionated) history of deep learning covering, up to 2020, its successes and investment by the gafam. It goes behind the scenes, focusing on the personalities who pushed the vision and engineering, the convincing of investors and the ideologies behind it. (2/2).
0
0
0
I recommend 'Genius makers' by Cade Metz, which could be renamed 'an enthusiastic history of the deep learning revolution'. (1/2).
1
0
5
RT @SecondmindLabs: Secondmind is opening five research placement positions in the area of Gaussian process models and Bayesian optimisatio….
0
17
0
RT @arnosolin: Our paper with @wil_j_wil and @vincentadam87 "Sparse Algorithms for Markovian Gaussian Processes" will be presented at #AIST….
0
6
0