Hi! I'm Aya Oshima, an AI engineer based in New York.
Selected Work
Thesis: Dimensional Reduction Techniques for Neural Data
Explored dimensional reduction techniques (PCA, TCA) applied to large-scale neural datasets. Conducted analysis on complex neural patterns, providing insights into brain activity and computational neuroscience, under the guidance of Prof. Robert C. Froemke.
Neural-Symbolic VQA
Designed a multi-modal AI system integrating computer vision, NLP, and symbolic reasoning to enhance visual question answering on the Sort-of-CLEVR dataset. Leveraged PyTorch to implement CNNs for image processing and integrated symbolic reasoning to provide interpretable results, improving accuracy and logical consistency.
Object Tracking Under Occlusions
Built a robust object tracking system combining CNN-based detection and Kalman filters to predict and track occluded objects effectively. Developed using ResNet for feature extraction and applied in scenarios requiring real-time, accurate tracking under challenging visual conditions.
Experience
Open Source Contributor
Python Typed APIs Project - Supervised by Prof. Kamen Yotov
- -Developing type-safe Python clients for widely-used services in an open-source project, automating API generation to address gaps in the software development industry.
Research Assistant
Neuroscience Kiani Lab, NYU Center for Neural Science
- -Contributed 5k+ eye-tracking data as a subject for the lab's decision-making experiments.
- -Developed a DeepDream algorithm using pre-trained CNNs (ResNet/VGG-16) to generate dream-like images, aiming for future use in neural decision-making experiments on primates.
- -Gained insights into research methodologies and expanded knowledge of neuroscience.
AI Robotics Engineer
Nihon Business Data Processing Center
- -Developed and deployed a voice-to-motion robotics system showcased to 10k+ exhibition visitors, leading the AI development as the sole AI engineer on a team of 5.
- -Integrated voice recognition (Julius/SRILM) and speech synthesis (JACK/OpenJTalk) into the robot, enabling seamless interaction between humans and robots.
- -Researched NLP for the future voice chatbot product, adjusting pre-trained models (BERT/GPT2/Blender), developing original small models (LSTM), and preprocessing messy 400kb data (MeCab/NLTK).
- -Played a project management role, leading cross-functional collaboration, timelines, and deployment strategies.