Duration: Fall and/or Spring 2023

Project Overview: The EpiNu project is developing lightweight digital strategies to tackle strategic gaps in the escalating climate health crisis. Guided by epigenetic and decolonial data insights, the current focus is on improving nutrition security for severely food-insecure women during preconception and prenatal stages, specifically in conflict and climate-affected areas, and with initial efforts in the Democratic Republic of Congo (DRC).

Job Summary: The selected undergraduate will focus on incremental learning and fine tuning to enhance accuracy and functionality of an existing image recognition model for an edge-fog-cloud data pipeline in low resource settings. 

Responsibilities Include:

Research & Literature Review:

• Conduct a comprehensive literature review on similar approaches across domains in low resource settings.
• Identify research gaps and propose new hypotheses for experimentation.

Explorative Data Analysis:

• Conduct in-depth analysis of the unique dataset to understand underlying structures, patterns, and potential biases.
• Identify missing, inconsistent, or anomalous data that may impact the research study’s integrity.

Data Cleaning and Preprocessing:

• Develop and apply robust techniques to handle missing or noisy data, ensuring that the dataset aligns with research standards.
• Collaborate with team to validate data cleaning methods, ensuring relevance to the research questions.
• Conduct data cleaning & preprocessing as needed and contribute to training of DRC field team for cleaning and preprocessing of image data.
• Augment data to improve model robustness.

Model Development & Innovation:

• Collaborate on incremental learning to adapt and update existing YOLO-based image recognition model with the team.
• Implement innovative techniques to enhance model performance and address nutrition security research questions.
• Fine tuning to enhance lightweight computation for edge deployment in low resource settings.
• Engage in iterative cycles of testing, learning, and refining the model with the US and DRC-based team.

Documentation Compliance:

• Maintain dataset documentation and research logs, including methodologies, experiments, results, and insights for US and DRC-based teams. 

Requirements: 

• Proficiency in YOLO architecture and related object detection techniques.
• Experience with deep learning frameworks like TensorFlow, PyTorch, or Keras.
• Strong programming skills in languages such as Python, C++, or Java.
• Experience in data preprocessing, cleaning, and augmentation.
• Knowledge of incremental learning techniques to adapt models to new data.
• Knowledge of optimizing models for efficiency, including low computational complexity and low latency.
• Strong analytical and problem-solving skills, with the ability to approach challenges creatively.

Please send your resume and statement of interest to blumcenter@berkeley.edu