publications

2025

1. ESPFormer: Doubly-Stochastic Attention with Expected Sliced Transport Plans

   Ashkan Shahbazi, Elaheh Akbari, Darian Salehi, and 3 more authors

   In Proceedings of the 42nd International Conference on Machine Learning (ICML), 2025

   Accepted to ICML 2025
2. Neural-Augmented Kelvinlet: Real-Time Soft Tissue Deformation with Multiple Graspers

   Ashkan Shahbazi, Kyvia Pereira, Jon S. Heiselman, and 12 more authors

   2025
3. Linear Spherical Sliced Optimal Transport: A Fast Metric for Comparing Spherical Data

   Xinran Liu, Yikun Bai, Rocı́o Dı́az Martı́n, and 5 more authors

   In Proceedings of the International Conference on Learning Representations (ICLR), 2025

   Accepted to ICLR 2025
4. Expected Sliced Transport Plans

   Xinran Liu, Rocı́o Dı́az Martı́n, Yikun Bai, and 4 more authors

   In Proceedings of the International Conference on Learning Representations (ICLR), 2025

   Accepted to ICLR 2025

2024

1. Understanding Learning with Sliced-Wasserstein Requires Rethinking Informative Slices

   Huy Tran, Yikun Bai, Ashkan Shahbazi, and 2 more authors

   2024
2. One Category One Prompt: Dataset Distillation using Diffusion Models

   Ali Abbasi, Ashkan Shahbazi, Hamed Pirsiavash, and 1 more author

   2024
3. Stereographic Spherical Sliced Wasserstein Distances

   Huy Tran, Yikun Bai, Abihith Kothapalli, and 4 more authors

   In Proceedings of the 41st International Conference on Machine Learning (ICML), 2024

   Accepted to ICML 2024
4. Equivariant vs. Invariant Layers: A Comparison of Backbone and Pooling for Point Cloud Classification

   Abihith Kothapalli, Ashkan Shahbazi, Xinran Liu, and 2 more authors

   In Proceedings of the ICML GRAM Workshop, 2024

   Accepted to the ICML GRAM Workshop 2024

2022

1. Detection of ADHD cases using CNN and classical classifiers of raw EEG

   Behrad TaghiBeyglou, Ashkan Shahbazi, Fatemeh Bagheri, and 2 more authors

   Computer Methods and Programs in Biomedicine Update, 2022

   Abs DOI

   Purpose: This study proposes a novel convolutional neural network (CNN) structure in conjunction with classical machine learning models, utilizing the raw electroencephalography (EEG) signal as the input to diagnose attention deficit hyperactivity disorder (ADHD) in children. The proposed EEG-based approach does not require transformation or artifact rejection techniques. Methods: In the first step, the suggested method uses raw EEG to train a CNN to diagnose ADHD. Then, the feature maps from different layers of the trained CNN are extracted and used to train some classical classifiers such as support vector machine (SVM), logistic regression (LR), random forest (RF), etc. This study benefits from an extended version of a dataset acquired from 61 participants diagnosed with ADHD and 60 individuals in control group, age 7 through 12 years old. Results: The initial CNN structure (without further use of feature maps) achieved an accuracy of 86.33±2.64% in 5-fold cross-validation scheme on training set, which is superior to results reported in previous studies. However, in order to increase the efficacy of the classifiers we used various feature representations across different CNN layers and after a rigorous evaluation of candidate classifiers, logistic regression provided an accuracy of 91.16±0.03% in training epochs using 5-fold cross-validation scheme and 95.83% in ADHD identification in unseen epochs, were achieved. Also, other metrics such as precision, sensitivity, F1-score and receiver of operating characteristic (ROC) were presented for better comparison of different hybrid methods. Conclusion: The suggested method for detection of ADHD in children shows high performance in different metrics such as accuracy, sensitivity, and specificity, which is superior to previously reported results.