no code implementations • 30 Mar 2023 • Thalaiyasingam Ajanthan, Matt Ma, Anton Van Den Hengel, Stephen Gould
In particular, it is necessary to circumvent the representational drift between the accumulated embeddings and the feature embeddings at the current training iteration as the learnable parameters are being updated.
no code implementations • ICCV 2023 • Peixia Li, Pulak Purkait, Thalaiyasingam Ajanthan, Majid Abdolshah, Ravi Garg, Hisham Husain, Chenchen Xu, Stephen Gould, Wanli Ouyang, Anton Van Den Hengel
Each learning group consists of a teacher network, a student network and a novel filter module.
no code implementations • 22 Dec 2022 • Kartik Gupta, Thalaiyasingam Ajanthan, Anton Van Den Hengel, Stephen Gould
Most current contrastive learning approaches append a parametrized projection head to the end of some backbone network to optimize the InfoNCE objective and then discard the learned projection head after training.
no code implementations • CVPR 2022 • Alexander Long, Wei Yin, Thalaiyasingam Ajanthan, Vu Nguyen, Pulak Purkait, Ravi Garg, Alan Blair, Chunhua Shen, Anton Van Den Hengel
We introduce Retrieval Augmented Classification (RAC), a generic approach to augmenting standard image classification pipelines with an explicit retrieval module.
Ranked #4 on Long-tail Learning on iNaturalist 2018
no code implementations • AAAI Workshop AdvML 2022 • Dishanika Dewani Denipitiyage, Thalaiyasingam Ajanthan, Parameswaran Kamalaruban, Adrian Weller
Lately, the literature on adversarial robustness spans from images to other domains such as point clouds.
no code implementations • 25 Mar 2021 • Amirreza Shaban, Amir Rahimi, Thalaiyasingam Ajanthan, Byron Boots, Richard Hartley
When the novel objects are localized, we utilize them to learn a linear appearance model to detect novel classes in new images.
1 code implementation • 9 Feb 2021 • Zhiwei Xu, Thalaiyasingam Ajanthan, Vibhav Vineet, Richard Hartley
In this work, we introduce a Resource Aware Neuron Pruning (RANP) algorithm that prunes 3D CNNs at initialization to high sparsity levels.
no code implementations • 1 Jan 2021 • Michele Sasdelli, Thalaiyasingam Ajanthan, Tat-Jun Chin, Gustavo Carneiro
Then, we empirically show that for a large range of learning rates, SGD traverses the loss landscape across regions with largest eigenvalue of the Hessian similar to the inverse of the learning rate.
1 code implementation • 9 Oct 2020 • Zhiwei Xu, Thalaiyasingam Ajanthan, Richard Hartley
We achieve it with fully-connected layers with Transparent Initialization (TI) and efficient logit consistency using a sparse encoder.
1 code implementation • 6 Oct 2020 • Zhiwei Xu, Thalaiyasingam Ajanthan, Vibhav Vineet, Richard Hartley
Specifically, the core idea is to obtain an importance score for each neuron based on their sensitivity to the loss function.
1 code implementation • ICLR 2021 • Kartik Gupta, Amir Rahimi, Thalaiyasingam Ajanthan, Thomas Mensink, Cristian Sminchisescu, Richard Hartley
From this, by approximating the empirical cumulative distribution using a differentiable function via splines, we obtain a recalibration function, which maps the network outputs to actual (calibrated) class assignment probabilities.
no code implementations • 23 Jun 2020 • Amir Rahimi, Thomas Mensink, Kartik Gupta, Thalaiyasingam Ajanthan, Cristian Sminchisescu, Richard Hartley
Calibration of neural networks is a critical aspect to consider when incorporating machine learning models in real-world decision-making systems where the confidence of decisions are equally important as the decisions themselves.
1 code implementation • 22 Jun 2020 • Yao Lu, Stephen Gould, Thalaiyasingam Ajanthan
The problem of vanishing and exploding gradients has been a long-standing obstacle that hinders the effective training of neural networks.
1 code implementation • 30 Mar 2020 • Kartik Gupta, Thalaiyasingam Ajanthan
In this work, we systematically study the robustness of quantized networks against gradient based adversarial attacks and demonstrate that these quantized models suffer from gradient vanishing issues and show a fake sense of robustness.
no code implementations • ICLR 2021 • Namhoon Lee, Thalaiyasingam Ajanthan, Philip H. S. Torr, Martin Jaggi
As a result, we find across various workloads of data set, network model, and optimization algorithm that there exists a general scaling trend between batch size and number of training steps to convergence for the effect of data parallelism, and further, difficulty of training under sparsity.
1 code implementation • ECCV 2020 • Amir Rahimi, Amirreza Shaban, Thalaiyasingam Ajanthan, Richard Hartley, Byron Boots
Weakly Supervised Object Localization (WSOL) methods only require image level labels as opposed to expensive bounding box annotations required by fully supervised algorithms.
1 code implementation • 24 Oct 2019 • Zhiwei Xu, Thalaiyasingam Ajanthan, Richard Hartley
In addition to differentiability, the two main aspects that enable learning these model parameters are the forward and backward propagation time of the MRF optimization algorithm and its inference capabilities.
1 code implementation • 18 Oct 2019 • Thalaiyasingam Ajanthan, Kartik Gupta, Philip H. S. Torr, Richard Hartley, Puneet K. Dokania
Quantizing large Neural Networks (NN) while maintaining the performance is highly desirable for resource-limited devices due to reduced memory and time complexity.
1 code implementation • ICLR 2020 • Namhoon Lee, Thalaiyasingam Ajanthan, Stephen Gould, Philip H. S. Torr
Alternatively, a recent approach shows that pruning can be done at initialization prior to training, based on a saliency criterion called connection sensitivity.
1 code implementation • CVPR 2019 • Alessio Tonioni, Oscar Rahnama, Thomas Joy, Luigi Di Stefano, Thalaiyasingam Ajanthan, Philip H. S. Torr
Real world applications of stereo depth estimation require models that are robust to dynamic variations in the environment.
6 code implementations • 27 Feb 2019 • Arslan Chaudhry, Marcus Rohrbach, Mohamed Elhoseiny, Thalaiyasingam Ajanthan, Puneet K. Dokania, Philip H. S. Torr, Marc'Aurelio Ranzato
But for a successful knowledge transfer, the learner needs to remember how to perform previous tasks.
Ranked #7 on Class Incremental Learning on cifar100
1 code implementation • ICCV 2019 • Thalaiyasingam Ajanthan, Puneet K. Dokania, Richard Hartley, Philip H. S. Torr
Compressing large Neural Networks (NN) by quantizing the parameters, while maintaining the performance is highly desirable due to reduced memory and time complexity.
no code implementations • 22 Nov 2018 • Richard Hartley, Thalaiyasingam Ajanthan
We consider move-making algorithms for energy minimization of multi-label Markov Random Fields (MRFs).
9 code implementations • ICLR 2019 • Namhoon Lee, Thalaiyasingam Ajanthan, Philip H. S. Torr
To achieve this, we introduce a saliency criterion based on connection sensitivity that identifies structurally important connections in the network for the given task.
no code implementations • 23 May 2018 • Thomas Joy, Alban Desmaison, Thalaiyasingam Ajanthan, Rudy Bunel, Mathieu Salzmann, Pushmeet Kohli, Philip H. S. Torr, M. Pawan Kumar
The presented algorithms can be applied to any labelling problem using a dense CRF with sparse higher-order potentials.
no code implementations • 17 Apr 2018 • Rodrigo de Bem, Arnab Ghosh, Thalaiyasingam Ajanthan, Ondrej Miksik, Adnane Boukhayma, N. Siddharth, Philip Torr
However, the latent space learned by such approaches is typically not interpretable, resulting in less flexibility.
2 code implementations • ECCV 2018 • Arslan Chaudhry, Puneet K. Dokania, Thalaiyasingam Ajanthan, Philip H. S. Torr
We observe that, in addition to forgetting, a known issue while preserving knowledge, IL also suffers from a problem we call intransigence, inability of a model to update its knowledge.
1 code implementation • CVPR 2016 • Thalaiyasingam Ajanthan, Richard Hartley, Mathieu Salzmann
Multi-label submodular Markov Random Fields (MRFs) have been shown to be solvable using max-flow based on an encoding of the labels proposed by Ishikawa, in which each variable $X_i$ is represented by $\ell$ nodes (where $\ell$ is the number of labels) arranged in a column.
no code implementations • CVPR 2017 • Thalaiyasingam Ajanthan, Alban Desmaison, Rudy Bunel, Mathieu Salzmann, Philip H. S. Torr, M. Pawan Kumar
To this end, we develop a proximal minimization framework, where the dual of each proximal problem is optimized via block coordinate descent.
no code implementations • CVPR 2015 • Thalaiyasingam Ajanthan, Richard Hartley, Mathieu Salzmann, Hongdong Li
While widely acknowledged as highly effective in computer vision, multi-label MRFs with non-convex priors are difficult to optimize.