Research Projects

AI can help consumers understand the health effects of their food beyond Nutrition Facts Labels. The goal is to empower users to know what they eat and how it affects their health. We will develop a "food photo to ingredients" and an "ingredients to health outcome" module. In collaboration with the Dietary Assistant project, these will integrate into a mobile app.

This project supports sustainable agriculture by using AI to reduce fertilizers, pesticides, and water while maintaining crop quality and yield through improved modeling and precision application. It focuses on soil and hydroponic crops. Part A covers modeling, simulation, and data integration, optimizing resource use, while Part B focuses on field data hardware.

This project tackles four bottlenecks in digital food safety across the farm-to-table supply chain by developing algorithms, data standards, and frameworks for privacy and text analytics. Focus areas include: 1) Active learning for monitoring optimization, 2) Standardized data for sharing, 3) Privacy-guaranteed data sharing, and 4) Machine learning for text analysis.

Our goal is to deliver healthy foods at low cost to improve nutrition for all Americans. Advances in chemistry reveal thousands of unique food molecules. This project extends ML models to include the gut microbiome and explores cost-health trade-offs in dietary choices.

The aims are to improve food safety and resilience in the poultry supply chain by: 1) Using machine learning to identify predictors of rare high-level Salmonella contamination in products, 2) Developing a federated learning approach for processors to privately improve safety, and 3) Modeling supply chain responses to contamination events to enhance AI-driven resilience.

This project is paired with a Part A. In Part B (this proposal) we consider the hardware capable of gathering the required data in the field, including sensors and robotics. The complementary Part A focuses on modeling and simulation for water, pesticides, and fertilizers in fields.

The project aims to develop AI models predicting starch digestibility in starchy foods (potatoes, rice) and assess the impact of clean label processing techniques. Objectives include: (a) AI models using imaging and DSC for individual foods, (b) AI models for starch digestibility in food-lipid systems, and (c) AI models integrating processing and product data.

We aim to develop and test novel, low-cost wireless sensors for measuring soil signals such as nitrate, ammonium, phosphate, potassium, and moisture. AI algorithms will predict the spatiotemporal distribution of these signals to support 4R Nutrient Stewardship.

This project aims to develop AgML, an open-source Python library to accelerate agricultural AI by providing access to ML datasets, benchmarks, pre-trained models, workflows, and synthetic data generators. It builds on the successes of the Year 3 AIFS proposal, "AgML: Open-Source Infrastructure to Accelerate Scaling of Agricultural AI Technologies."

To address fresh produce waste and spoilage, we propose AI-based predictive solutions: (a) rapid detection and quantification of spoilage microbes in low-resource settings, (b) prediction of spoilage risks based on ripening, microbes, and transport, and (c) semi-supervised domain generalization for these aims, integrating Aims 1 and 2 for data-efficient AI models.

Our work identifies key stakeholders and critics of AI technologies, learns their perspectives on core AI and food system issues, and examines points of resistance (e.g., hesitation, legal, logistical, or ethical barriers) and relational tensions. We aim to reframe problems from their viewpoints to ease tensions, foster dialogue, and build pathways forward.

The goal of this project is to develop a dietary assistant using a large language model linked to a food-to-chemical knowledge graph. It will offer dietary recommendations, allergen identification, and nutritional information. Specific aims include developing a chatbot with FoodAtlas integration, adding an explainability module, and creating a personalized dietician agent.

This project aims to (a) create a pipeline for screening peptides in food related to antimicrobial, antiflammatory, and obesity-related activities, (b) develop predictors for biochemical pathways that generate bioactive compounds influencing human health and sensory from various chemical compounds.

Biofortification tackles global micronutrient deficiencies through crop improvement without added inputs. To optimize breeding programs, an AI toolkit is being developed to predict and evaluate nutritional quality using machine learning. Open-source resources will benefit broader initiatives, and a hackathon will engage UC Davis students in advancing agricultural innovation.

Utilizing AI, researchers aim to enhance real-time dietary assessment using photo diaries. The project's long-term goal is to develop modules linking food photos to ingredients and ingredients to health outcomes, incorporating novel data types like the glycome of food and hierarchical dietary representations for improved predictions.

Biofortification tackles global micronutrient deficiencies through crop improvement without added inputs. To optimize breeding programs, an AI toolkit is being developed to predict and evaluate nutritional quality using machine learning. Open-source resources will benefit broader initiatives, and a hackathon will engage UC Davis students in advancing agricultural innovation.

Food digestion is crucial for nutrient release and macromolecule breakdown, yet the absence of predictive models hinders health and innovation in the food industry. This project seeks to fill this gap, aiming to develop AI technologies for improved food manufacturing, personalized nutrition, and digital twin models of gut and food digestion.

This UC Berkeley and UC Davis project develops low-cost wireless soil sensors and AI algorithms, tested in controlled settings and commercial fields using a UAV-equipped wireless readout system, with the goal of delivering affordable real-time soil monitoring tools for improved agricultural practices and environmental stewardship.

Recent studies and commercial applications focus on automated food identification for calorie and macronutrient estimation. However, existing models trained on internet-based recipes may struggle in real scenarios, prompting the project to develop an explainable diet recommendation system that considers fine-grained information like ingredients for accurate dietary analysis.

Researchers investigates the potential of data-driven innovation in achieving digital "smart" food safety by interviewing industry leaders about sharing safety data. The team will then create curated datasets and develop digital twins for food systems which will integrate into a comprehensive retail food safety and spoilage model for dynamic risk assessment.

The project aims to build infrastructure for the rapid scaling of agricultural AI technologies by creating an open-source Python library, AgML, providing access to agricultural-specific machine learning datasets, benchmarks, pre-trained models, workflows, and synthetic data generators, building on successes from the Year 2 AIFS proposal.

AIFS sponsored FoodAtlas, an automated knowledgebase of food chemical composition from 2000 papers, covering 281 foods and 1268 chemicals. The team plans to expand the AI methods to curate additional entity types, open FoodAtlas with APIs and user interface, and make an educational module on bioactive molecules in food to accelerate research and commercialization applications.

The project targets the isolated nature of data in food processing, by fostering collaboration and developing AI infrastructure in areas like food safety, Federated Learning, and inventory strategies. The broader impact is the creation of a food safety testing database for public-domain data, addressing the need for data availability and becoming a widely used exploration tool.

The project aims to develop digital twin and machine learning algorithms for optimizing LED-driven pod-based hydroponic farming, with a focus on addressing plant pathogen challenges, offering potential impacts on predicting pathogen spread, decontamination, simulating hydroponic systems, and revolutionizing the food processing industry if successful.

After gathering insights on AIFS processes and farmers' challenges, the researchers expand their work by refining an ethics framework, broadening stakeholder engagement through interviews with various groups, and assessing the economic benefits of applying advanced technology, contributing to AIFS's mission of advancing AI infrastructure and supporting healthy food systems.

This project advances on hydroponic farming for sustainable large-scale indoor agriculture by developing digital-twin technologies, with an emphasis on training professionals in engineering, computer science, and agriculture, and addressing food security and pathogen control challenges through innovative solutions like LED-driven pod-based hydroponic farming systems.

The overall goal is to develop Digital Twin and Machine Learning algorithms for addressing key food safety challenges associated with introduction and spread of pathogens in food facilities and enable guided decontamination for food contact surfaces to reduce risk of cross-contamination.

The project involves collaboration on open-source infrastructure development, optimized sensor-driven precision agriculture, and inexpensive soil nitrate sensors, with the potential to accelerate agricultural AI technologies through AgML Python library and to develop tools for autonomous vehicles and soil nitrate measurement.

This project aims to utilize AI in pioneering a pilot initiative to collect data for targeted dietary interventions, assessing current food distributions, developing predictive models, mapping out food bank processes, creating a pilot AI solution for personalized food distribution, and establishing a platform to measure the impact of food interventions on food bank clients.

The project focuses on accelerating the scaling of agricultural AI technologies by creating an open-source Python library, AgML, consolidating machine learning datasets, benchmarks, pre-trained models, workflows, and synthetic data generators, building on successes from two Year 1 AIFS proposals and aligning with the Year 2 RFP.

Researches propose these actions to support AIFS in achieving its mission by establishing a meaningful and viable ethical foundation that will support our researchers in carrying out their work, and support trust among those who will ultimately use AIFS technologies.

The proposed multidisciplinary project will develop a framework for an integrated smarter food safety that achieves a vertical integration of AI systems for food safety across the supply chain in order to predict and assess the food safety risk in the food product at consumption (at the end of the supply chain) based on a holistic view of the entire supply chain.

The proposed research aims to fulfill an unmet need in precision nutrition and additive manufacturing of foods by developing AI-enabled predictive models for 3D printed foods and a gut digestion process to optimize the bioaccessibility of micronutrients.

This two-part project aims to develop predictors for nutritional quality and aroma by leveraging genetic and phenotypic features, employing machine learning to assess the overall value of new varieties based on their molecular compositions for flavor and health effects, specifically focusing on three major compound classes: phenols, terpenes, and peptides.

AIFS is collaborating with UC Davis to comprehensively analyze the composition of milk using various 'omics analyses, aiming to create a database that connects molecular information with research-derived data on nutritional and health benefits, facilitating a deeper understanding of the health effects of milk molecules through machine learning-derived insights.

The overall impact of this work is to accelerate knowledge discovery in health and nutrition by establishing essential infrastructure, such as databases and algorithms, while specifically aiming to create and refine an end-to-end diet recommendation system with a proof-of-concept implementation, contributing to a healthier society through improved food choices.

In systems engineering, the adoption of advanced AI approaches, particularly the integration of digital twins for adaptive system operation with human involvement, is progressing rapidly. But AI approaches in food production systems lags behind. The development of tools for food systems researchers aims to enhance research capabilities and contribute to global benefits.

The AI Institute for Next Generation Food Systems aims to develop AI tools for transforming US food systems, emphasizing the need for a clear ethical framework, proposing two projects to establish recommendations for trustworthy research practices and create an ethics curriculum for researchers and students involved in agricultural applications.

The research investigates AI-guided ultrasonic cleaning of produce, with a focus on enhancing cleaning effectiveness through shear induced on the produce surface by bubble-enhanced ultrasonic cleaning, spanning a range of amplitudes from oscillating non-condensing gas bubbles to cavitation near an artificial leaf surface designed to mimic real lettuce leaf characteristics.

The project aims to address the challenges in assessing the effects of diet on human health by developing AI-assisted methods for real-time capture of diet and predicting the glycan content in images of mixed meals from real-world food diaries to enhance understanding of the impact on the gut microbiome.

The project's overarching goal is to create a digital twin and machine-learning optimization framework for rapid path planning in real-time, addressing fluid dynamics, vehicle dynamics, energy-efficient path planning, and data extraction for agricultural mapping in large surface area food systems.

The project aims to advance non-destructive sensing in leafy greens and legume breeding programs by leveraging hyperspectral imaging and AI methods to predict plant physiological traits, nutrient uptake, water/carbon use, and canopy architecture.

The project addresses the food processing industry's need for food-safety data analytics aiming to reduce contamination risk by developing environmental monitoring programs through agent-based simulation models to enhance decision-making in complex situations and under stress, overcoming challenges that hinder their full utilization.

An AI regression model will be developed using 10 years of data from the Processing Tomato Advisory Board and a digital model of mold growth to predict tomato mold severity, likelihood of load rejection, and spoilage, with the aim of improving the accuracy of predicting tomato damage and spoilage in processing facilities.

Efforts to expand food composition information, exemplified by FoodDB, encounter challenges of manual curation and the presence of "nutritional dark matter," prompting a proposed solution involving the creation of a software package for automated curation of nutrition data from papers.

Enhancing yield forecasting for specialty crops, such as strawberries, grapes, and almonds, involves deploying monitoring sensors, developing a synthetic data generation pipeline linked to a 3D crop model and deep learning, and creating a framework that fuses multi-modal ground-based and aerial data to predict yield.

Addressing challenges in data efficiency when developing effective AI algorithms for next-generation food systems involve identifying key issues in molecular breeding, agriculture, food processing, and nutrition, and developing active-learning and sim-to-real approaches to enhance data efficiency in machine learning algorithms for food systems.

Integrating Performance-Based Engineering with AI in a food system management framework, the study evaluates the performance of existing food facilities, assesses methods for enhancing crop yield and controlling diseases, and quantifies the benefits of real-time monitoring and automated machinery on the resilience and performance of food facilities under environmental hazards.

This study creates a simulation tool to quantify and visualize diverse outcomes of selection strategies across multiple traits and allocate phenotyping resources for improved genetic value prediction accuracy, utilizing Sequential Design of Experiments principles in the context of the UC Davis strawberry and pepper breeding programs.

Addressing food safety concerns related to pathogen transmission through air and surfaces, the development of digital twin and machine learning algorithms can enhance mitigation and treatment strategies, particularly in the context of thin biofilms on food processing surfaces.

The proposed XRL framework that utilizes reinforcement learning addresses control and decision-making challenges in agriculture, food processing/distribution, and molecular breeding by accessing a simulator/model of the food system and relevant variables provided by domain experts.

The project aims to develop a data-efficient, AI-enabled framework for water and nitrogen stress sensing and prediction on almond. Having rapid detection and prediction of water and nitrogren stress is crucial for optimizing production and enabling precision management activities to enhance grower output and quality while minimizing negative environmental impacts.