Research

PRISM: Reducing Spurious Implicit Biases in Vision-Language Models with LLM-Guided Embedding Projection

Mahdiyar Molahasani^*, Azadeh Motamedi^*, Michael Greenspan, Il-Min Kim, Ali Etemad

^* Equal contribution

• VLM
• Bias & Fairness
• CLIP
• ICCV

PRISM introduces a data-free, task-agnostic debiasing framework for VLMs. It first leverages an LLM to generate bias-aware scene descriptions from simple class prompts, then learns a linear projection of the CLIP embedding space via a novel Latent-space Debiasing loss that enforces intra-class invariance and inter-class separability.

Federated Unsupervised Domain Generalization using Global and Local Alignment of Gradients

Farhad Pourpanah^*, Mahdiyar Molahasani^*, Milad Soltany^*, Ali Etemad, Michael Greenspan

^* Equal contribution

• Federated Learning
• Domain Generalization
• AAAI

We introduced the novel problem of unsupervised federated domain generalization and proposed FedGaLA, a method that improves model generalization across unseen domains by aligning gradients at both the client and server levels. This work is grounded in a theoretical framework that links domain shift to gradient alignment. FedGaLA achieves state-of-the-art performance on several domain generalization benchmarks.

Federated Domain Generalization With Label Smoothing and Balanced Decentralized Training

Milad Soltany^*, Farhad Pourpanah^*, Mahdiyar Molahasani^*, Michael Greenspan, Ali Etemad

^* Equal contribution

• Federated Learning
• Domain Generalization
• ICASSP

We propose FedSB, a method for federated domain generalization that improves model robustness across diverse domains using label smoothing to reduce local overconfidence and a balanced training mechanism to mitigate data heterogeneity.

AI-powered Low-order Focal Plane Wavefront Sensing in Infrared

Mojtaba Taheri, Mahdiyar Molahasani, Sam Ragland, Benoit Neichel, Peter Wizinowich

• Astronomy
• Adaptive Optics
• Applied AI

We propose an AI-powered FPWFS method specifically for low-order mode estimation in infrared wavelengths. Our approach is trained on simulated data and validated on on-telescope data collected from the Keck I adaptive optic (K1AO) bench calibration source in K-band, paving the way for more compact AO systems for astronomical observations.

Automated Tooth Extraction and Caries Detection

Arman Haghanifar, Mahdiyar Molahasani, Seok-bum Ko

• Medical Imaging
• Dental
• Capsule Networks

A fully automated tooth extraction model is implemented using a genetic algorithm. A multi-feature extraction model with a capsule classifier is developed for caries detection.

Continual Learning for Long-Tailed Recognition

Mahdiyar Molahasani, Ali Etemad, Michael Greenspan

• Continual Learning
• Long-Tailed
• NeurIPS

This work presents a theoretical framework for addressing long-tailed recognition (LTR) through continual learning (CL), where models are trained sequentially on data subsets to balance performance across head (frequent) and tail (rare) classes. By proving bounds on model weight updates and demonstrating CL's effectiveness on benchmark datasets, we show that CL can significantly improve LTR performance.

Continual Learning for Out-of-Distribution Generalization in Pedestrian Detection

Mahdiyar Molahasani, Ali Etemad, Michael Greenspan

• Continual Learning
• Object Detection
• OOD
• ICIP

This study introduces the first continual learning approach for pedestrian detection that can effectively address distribution shift, a common issue in prior works. We proposed modified Elastic Weight Consolidation for object detection networks, enabling the model to maintain its performance across different datasets and significantly improve the miss rate on CrowdHuman and CityPersons datasets by mitigating catastrophic forgetting.

Multi-scale Multi-task Crowd Counting

Mohsen Zand, Haleh Damirchi, Andrew Farley, Mahdiyar Molahasani, Michael Greenspan, Ali Etemad

• Crowd Counting
• Multi-task
• ICASSP

A multi-scale crowd counting and localization platform is proposed in this work. This novel architecture alongside the multi-scale multi-task loss function has demonstrated promising performance on standard benchmarks.

COVID-CXNet: Detecting COVID-19 in Frontal Chest X-ray Images

Arman Haghanifar, Mahdiyar Molahasani, Younhee Choi, S. Deivalakshmi, Seok-bum Ko

• Medical Imaging
• COVID-19
• Open Dataset

In this work, the largest publicly available dataset for COVID-19 chest X-rays is collected and a powerful COVID-19 detection model based on CheXNet is proposed. The model can detect COVID-19 accurately using meaningful features verified via class-activation maps.

Erosion Detection in Hydraulic Tubes and Hoses Using GRU

Elnaz Etminan, Mahdiyar Molahasani, Seok-bum Ko, Travis Wiens

• Time Series
• GRU
• Industrial

The characteristics of the eroded area in the pipe are extracted from the pressure response using a GRU network. This work represents the first erosion detection system leveraging deep learning.

High-scale Prostate MRI Super-Resolution with MSG-CapsGAN

Mahdiyar Molahasani, Younhee Choi, S. Deivalakshmi, Seok-bum Ko

• Medical Imaging
• Super-Resolution
• GAN

One of the first attempts for high-scale super-resolution (8x) in the biomedical domain. MSG-CapsGAN shows promising results in the medical domain as well.

Hybrid CMOS/Memristor Crossbar Implementation of Recurrent Neural Networks

Mahdiyar Molahasani, Jafar Shamsi, S. B. Shokouhi, Seok-bum Ko

• Hardware AI
• Memristor
• RNN

An efficient and scalable transistor-level implementation of two different recurrent neural networks is proposed using a memristor crossbar array.

MSG-Caps GAN for Face Super-Resolution

Mahdiyar Molahasani, Seok-bum Ko

• Super-Resolution
• GAN
• Capsule Networks

We proposed the first Multi-scale gradient capsule GAN and utilized it for face super-resolution. This model outperformed state-of-the-art face super-resolution models at high upscaling factors.

Anomaly Prediction in 5G Network

Ramin Sharifi, Mahdiyar Molahasani, Vahid Tabataba Vakili

• Time Series
• LSTM
• Telecom

An LSTM network is utilized for user activity prediction in 5G networks. The proposed model can accurately predict anomalies up to one hour in advance.