no code implementations • 8 Mar 2024 • Amitis Shidani, Devon Hjelm, Jason Ramapuram, Russ Webb, Eeshan Gunesh Dhekane, Dan Busbridge
Contrastive learning typically matches pairs of related views among a number of unrelated negative views.
no code implementations • 6 Dec 2023 • Polina Turishcheva, Jason Ramapuram, Sinead Williamson, Dan Busbridge, Eeshan Dhekane, Russ Webb
Understanding model uncertainty is important for many applications.
no code implementations • 7 Sep 2023 • Skyler Seto, Barry-John Theobald, Federico Danieli, Navdeep Jaitly, Dan Busbridge
In online F-TTA, a pre-trained model is adapted using a stream of test samples by minimizing a self-supervised objective, such as entropy minimization.
1 code implementation • 20 Jul 2023 • Borja Rodríguez-Gálvez, Arno Blaas, Pau Rodríguez, Adam Goliński, Xavier Suau, Jason Ramapuram, Dan Busbridge, Luca Zappella
We consider a different lower bound on the MI consisting of an entropy and a reconstruction term (ER), and analyze the main MVSSL families through its lens.
1 code implementation • 28 Jun 2023 • Xavier Suau, Federico Danieli, T. Anderson Keller, Arno Blaas, Chen Huang, Jason Ramapuram, Dan Busbridge, Luca Zappella
We propose 2D strUctured and EquivarianT representations (coined DUET), which are 2d representations organized in a matrix structure, and equivariant with respect to transformations acting on the input data.
1 code implementation • 11 Mar 2023 • Shuangfei Zhai, Tatiana Likhomanenko, Etai Littwin, Dan Busbridge, Jason Ramapuram, Yizhe Zhang, Jiatao Gu, Josh Susskind
We show that $\sigma$Reparam provides stability and robustness with respect to the choice of hyperparameters, going so far as enabling training (a) a Vision Transformer {to competitive performance} without warmup, weight decay, layer normalization or adaptive optimizers; (b) deep architectures in machine translation and (c) speech recognition to competitive performance without warmup and adaptive optimizers.
no code implementations • 28 Oct 2022 • Andrius Ovsianas, Jason Ramapuram, Dan Busbridge, Eeshan Gunesh Dhekane, Russ Webb
Self-supervised representation learning (SSL) methods provide an effective label-free initial condition for fine-tuning downstream tasks.
1 code implementation • 15 Jul 2022 • Shuangfei Zhai, Navdeep Jaitly, Jason Ramapuram, Dan Busbridge, Tatiana Likhomanenko, Joseph Yitan Cheng, Walter Talbott, Chen Huang, Hanlin Goh, Joshua Susskind
This pretraining strategy which has been used in BERT models in NLP, Wav2Vec models in Speech and, recently, in MAE models in Vision, forces the model to learn about relationships between the content in different parts of the input using autoencoding related objectives.
no code implementations • 1 Oct 2021 • Tom George Grigg, Dan Busbridge, Jason Ramapuram, Russ Webb
Despite the success of a number of recent techniques for visual self-supervised deep learning, there has been limited investigation into the representations that are ultimately learned.
no code implementations • 1 Oct 2021 • Jason Ramapuram, Dan Busbridge, Xavier Suau, Russ Webb
While state-of-the-art contrastive Self-Supervised Learning (SSL) models produce results competitive with their supervised counterparts, they lack the ability to infer latent variables.
no code implementations • 1 Oct 2021 • Jason Ramapuram, Dan Busbridge, Russ Webb
In this work we examine how fine-tuning impacts the fairness of contrastive Self-Supervised Learning (SSL) models.
1 code implementation • 27 Jul 2020 • Joseph Enguehard, Dan Busbridge, Adam Bozson, Claire Woodcock, Nils Y. Hammerla
The modelling of Electronic Health Records (EHRs) has the potential to drive more efficient allocation of healthcare resources, enabling early intervention strategies and advancing personalised healthcare.
no code implementations • 28 Mar 2020 • Albert Buchard, Baptiste Bouvier, Giulia Prando, Rory Beard, Michail Livieratos, Dan Busbridge, Daniel Thompson, Jonathan Richens, Yuanzhao Zhang, Adam Baker, Yura Perov, Kostis Gourgoulias, Saurabh Johri
We show that this approach is on a par with human performance, yielding safe triage decisions in 94% of cases, and matching expert decisions in 85% of cases.
no code implementations • 4 Oct 2019 • Joseph Enguehard, Dan Busbridge, Vitalii Zhelezniak, Nils Hammerla
The choice of sentence encoder architecture reflects assumptions about how a sentence's meaning is composed from its constituent words.
2 code implementations • ICLR 2019 • Dan Busbridge, Dane Sherburn, Pietro Cavallo, Nils Y. Hammerla
We investigate Relational Graph Attention Networks, a class of models that extends non-relational graph attention mechanisms to incorporate relational information, opening up these methods to a wider variety of problems.
1 code implementation • ICLR 2018 • Vitalii Zhelezniak, Dan Busbridge, April Shen, Samuel L. Smith, Nils Y. Hammerla
Experimental evidence indicates that simple models outperform complex deep networks on many unsupervised similarity tasks.