Laknath Ashwin De Silva

Hello there! My name is Laknath Ashwin De Silva (ලක්නාත් අශ්වින් ද සිල්වා in Sinhalese script). I am a second-year PhD student in the Department of Biomedical Engineering at Johns Hopkins University, where I am fortunate to be advised by Dr. Joshua Vogelstein and co-advised by Dr. Pratik Chaudhari (UPenn) and Dr. Carey E. Priebe (JHU).

I broadly work on machine learning, with an aspiration to strive towards reducing the gap between artificial and natural intelligences. My doctoral research is currently focused on out-of-distribution (OOD) generalization, robustness to distribution shifts, and domain adaptation. Additionally, I am interested in causal inference, generative modeling, cognitive neuroscience, and biomedical data science. In the past, I have worked on a number of computer vision and signal processing problems.

I received my Bachelor’s degree in Biomedical Engineering from University of Moratuwa, Sri Lanka in 2020 ( ranked 1st among ~950 undergrads in the Faculty of Engineering). I completed my thesis on “Designing a Cost-Effective Dry Contact sEMG Sensor System for Controlling a Bionic Hand” with Dr. Simon Kappel and Dr. Thilina Lalitharatne.

I have worked as a Research Assistant at the Florey Institute of Neuroscience and Mental Health, University of Melbourne, Australia in 2017 and 2018 under the supervision of Prof. Steve Petrou and Prof. Saman Halgamuge. There, I developed ML & Signal Processing software for modelling genetic epilepsy using high density multi-electrode array (MEA) signals recorded from in-vitro neuronal networks.

I have worked with Dr. Steffen Bollmann (Center for Advanced Imaging, University of Queensland, Australia) on developing fully convolutional networks to unwrap wrapped phase in QSM images.

Outside work, I love spending time with my wife Malsha, playing the piano, hiking, and trying out new recipes. I am an avid fan of astronomy, history, science-fiction (Star Wars!), fantasy, and anime!

news

Oct 23, 2022	Our work on “The Value of Out-of-Distribution Data” won the best short paper award at the OOD-CV workshop, ECCV 2022! 🏆
Oct 20, 2022	1 Paper accepted to the NeurIPS 2022 workshop on Distribution Shift, New Orleans, LA! 💫
Aug 22, 2022	Our paper titled “The Value of Out-of-Distribution Data” got accepted to the OOD Generalization in Computer Vision workshop at ECCV 2022, Tel Aviv, Israel! 💫
Jul 17, 2022	Attended ICML 2022, Baltimore, MD in person.
Jun 19, 2022	Attended CVPR 2022, New Orleans, LA in person.

selected publications

[W] NeurIPS’22

The Value of Out-of-Distribution Data

De Silva, Ashwin*, Ramesh, Rahul*, Priebe, Carey E, Chaudhari, Pratik, and Vogelstein, Joshua T

In NeurIPS 2022 Workshop on Distribution Shifts 2022

Abs arXiv

More data helps us generalize to a task. But real datasets can contain out-of-distribution (OOD) data; this can come in the form of heterogeneity such as intra-class variability but also in the form of temporal shifts or concept drifts. We demonstrate a counter-intuitive phenomenon for such problems: generalization error of the task can be a non-monotonic function of the number of OOD samples; a small number of OOD samples can improve generalization but if the number of OOD samples is beyond a threshold, then the generalization error can deteriorate. We also show that if we know which samples are OOD, then using a weighted objective between the target and OOD samples ensures that the generalization error decreases monotonically. We demonstrate and analyze this issue using linear classifiers on synthetic datasets and medium-sized neural networks on CIFAR-10.
[W] ECCV’22

The Value of Out-of-Distribution Data

De Silva, Ashwin*, Ramesh, Rahul*, Priebe, Carey E, Chaudhari, Pratik, and Vogelstein, Joshua T

In ECCV 2022 Workshop on Out-of-distribution Generalization in Computer Vision 2022

Abs arXiv HTML

More data helps us generalize to a task. But real datasets can contain out-of-distribution (OOD) data; this can come in the form of heterogeneity such as intra-class variability but also in the form of temporal shifts or concept drifts. We demonstrate a counter-intuitive phenomenon for such problems: generalization error of the task can be a non-monotonic function of the number of OOD samples; a small number of OOD samples can improve generalization but if the number of OOD samples is beyond a threshold, then the generalization error can deteriorate. We also show that if we know which samples are OOD, then using a weighted objective between the target and OOD samples ensures that the generalization error decreases monotonically. We demonstrate and analyze this issue using linear classifiers on synthetic datasets and medium-sized neural networks on CIFAR-10.
[C] ICASSP’22

Towards Accurate Cross-Domain in-Bed Human Pose Estimation

Afham, Mohamed, Haputhanthri, Udith, Pradeepkumar, Jathurshan, Anandakumar, Mithunjha, De Silva, Ashwin, and Edussooriya, Chamira U. S.

In ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) 2022

Abs HTML

Human behavioral monitoring during sleep is essential for various medical applications. Majority of the contactless human pose estimation algorithms are based on RGB modality, causing ineffectiveness in in-bed pose estimation due to occlusions by blankets and varying illumination conditions. Long-wavelength infrared (LWIR) modality based pose estimation algorithms overcome the aforementioned challenges; however, ground truth pose generations by a human annotator under such conditions are not feasible. A feasible solution to address this issue is to transfer the knowledge learned from images with pose labels and no occlusions, and adapt it towards real world conditions (occlusions due to blankets). In this paper, we propose a novel learning strategy comprises of two-fold data augmentation to reduce the cross-domain discrepancy and knowledge distillation to learn the distribution of unlabeled images in real world conditions. Our experiments and analysis show the effectiveness of our approach over multiple standard human pose estimation baselines.
[C] ICASSP’21

A Joint Convolutional and Spatial Quad-Directional LSTM Network for Phase Unwrapping

Perera, Malsha V.*, and De Silva, Ashwin*

In ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) 2021

Abs HTML PDF Code Video

Phase unwrapping is a classical ill-posed problem which aims to recover the true phase from wrapped phase. In this paper, we introduce a novel Convolutional Neural Network (CNN) that incorporates a Spatial Quad-Directional Long Short Term Memory (SQD-LSTM) for phase unwrapping, by formulating it as a regression problem. Incorporating SQD-LSTM can circumvent the typical CNNs’ inherent difficulty of learning global spatial dependencies which are vital when recovering the true phase. Furthermore, we employ a problem specific composite loss function to train this network. The proposed network is found to be performing better than the existing methods under severe noise conditions (Normalized Root Mean Square Error of 1.3% at SNR = 0 dB) while spending a significantly less computational time (0.054s). The network also does not require a large scale dataset during training, thus making it ideal for applications with limited data that require fast and accurate phase unwrapping.
[C] ICASSP’20

Real-Time Hand Gesture Recognition Using Temporal Muscle Activation Maps of Multi-Channel sEMG Signals

De Silva, Ashwin*, Perera, Malsha V.*, Wickramasinghe, Kithmin, Naim, Asma M., Dulantha Lalitharatne, Thilina, and Kappel, Simon L.

In ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) 2020

Abs HTML PDF Code Video

Accurate and real-time hand gesture recognition is essential for controlling advanced hand prostheses. Surface Electromyography (sEMG) signals obtained from the forearm are widely used for this purpose. Here, we introduce a novel hand gesture representation called Temporal Muscle Activation (TMA) maps which captures information about the activation patterns of muscles in the forearm. Based on these maps, we propose an algorithm that can recognize hand gestures in real-time using a Convolution Neural Network. The algorithm was tested on 8 healthy subjects with sEMG signals acquired from 8 electrodes placed along the circumference of the forearm. The average classification accuracy of the proposed method was 94%, which is comparable to state-of-the-art methods. The average computation time of a prediction was 5.5ms, making the algorithm ideal for the real-time gesture recognition applications.