๐๐ผ Hi, Iโm Adrian!
๐ Iโm a final year student at the University of Texas at Dallas pursuing a B.S. in Computer Science with a specialization in Artificial Intelligence and Machine Learning. Iโm an aspiring Ph.D. candidate passionate about making AI more accessible through efficient and optimized architectures.
๐ต๐ผ My research interests lie at the intersection of neural network efficiency and theoretical understanding:
- Neural network pruning and optimization
- Double descent phenomenon in deep learning
- Superposition in large neural networks
- Topological data analysis in deep learning
- Few-shot learning and vision transformers
๐จ๐ฝโ๐ป Iโm particularly fascinated by how modifying parameter spaces impacts architecture performance and generalization. My work spans computer vision, topological machine learning, and multimodal architectures.
Current Research
๐ฌ Audio-Visual Transformer Benchmarking
Working with Dr. Yunhui Guo on developing rigorous benchmarks for multimodal transformers, with a focus on safety-critical applications like autonomous vehicles. Creating comprehensive evaluation frameworks using AudioSet, Kinetics, and VGGSound datasets. Currently developing novel architectures to better handle audio-visual fusion and improve performance on these benchmarks.
๐ฏ Few-Shot Learning Enhancement
Working with Dr. Yunhui Guo and Dr. Baris Coskunuzer on improving cross-domain few-shot learning through topological data analysis and transformer architectures. Currently developing tuned encoder for Betti vector implementations for enhanced classification tasks using the SwinV2 backbone. Code
๐ฅ Human Identification in Videos
Collaborating with Dr. Alice OโToole on enhancing human identification in video sequences using modified vision transformers, focusing on improving generalization and accuracy by using the SwinV2 transformer instead of the current leading ViT.
๐งฎ Statistical Pruning of Neural Networks
Working with Dr. Richard Golden at the COINS Lab on developing novel pruning algorithms using sensitivity-based statistical methods. Lead developer of StatPruneNet, a benchmark toolkit for evaluating statistical pruning algorithms.
Previous Research
๐ Topological Vision Transformers
Collaborated with Dr. Baris Coskunuzerโs Topological Machine Learning Group on integrating topological data analysis with vision transformers, in particular SwinV2 transformer. Developed a novel approach achieving 10% improvement in breast cancer detection compared to baseline models and over 20% compared to topological CNNs using cross-attention between topological data analysis transformer encoders. Code
Teaching & Mentoring
๐จโ๐ซActive member of the Society for Advancement of Chicanos/Hispanics and Native Americans in Science, dedicated to mentoring underrepresented students in STEM fields.
Technical Skills
๐ป Programming: Python, C/C++, R, MATLAB
๐ค ML/DL: PyTorch, TensorFlow, HuggingFace, JAX
๐ ๏ธ Development: CUDA, Git, Linux/Unix
๐ Scientific: NumPy, Pandas, Matplotlib