Ali Ahmadi Roghabadi

Skin Lesion Classification — This project is a lightweight, multi-modal deep learning model that classifies skin lesions from images into seven diagnostic categories from the HAM10000 dataset — melanoma, melanocytic nevi, basal cell carcinoma, actinic keratoses/Bowen's disease, benign keratosis-like lesions, dermatofibroma, and vascular lesions — while also taking the patient's age, sex, and lesion body location as additional inputs to make more clinically informed predictions than an image-only model. Built on a MobileNetV3 backbone, the model is designed with a deliberate focus on accuracy and efficiency alongside small resource usage rather than raw benchmark performance alone, making it lightweight enough to embed directly into mobile apps and run inference on-device, without depending on a server or cloud API for every prediction.

LIF Neuron Simulation — Implementation and simulation of a single leaky integrate-and-fire (LIF) neuron, one of the simplest and most widely used models in computational neuroscience for approximating how a biological neuron behaves. The neuron's membrane potential is modeled as it integrates incoming input current over time while also "leaking" charge back toward a resting state, mimicking the passive electrical properties of a real neuron's membrane. once the membrane potential crosses a defined threshold, the model generates a spike and resets the potential, reproducing the basic firing behavior seen in real neurons. The project visualizes this membrane potential dynamics and spike generation over time.

Whisper Fine-tuning — Fine-tuned OpenAI's Whisper model on Persian speech datasets to enhance automatic speech recognition (ASR) performance. This project was conducted at the Deep Learning Lab, University of Tehran.