Food waste and insecurity are two societal challenges that coexist in many parts of the world. A prominent force to combat these issues, food rescue platforms match food donations to organizations that serve underprivileged communities, and then rely on external volunteers to transport the food. Previous work has developed machine learning models for food rescue volunteer engagement. However, having long worked with domain practitioners to deploy AI tools to help with food rescues, we understand that there are four main pain points that keep such a machine learning model from being actually useful in practice: small data, data collected only under the default intervention, unmodeled objectives due to communication gap, and unforeseen consequences of the intervention. In this paper, we introduce bandit data-driven optimization which not only helps address these pain points in food rescue, but also is applicable to other nonprofit domains that share similar challenges. Bandit data-driven optimization combines the advantages of online bandit learning and offline predictive analytics in an integrated framework. We propose PROOF, a novel algorithm for this framework and formally prove that it has no-regret. We show that PROOF performs better than existing baseline on food rescue volunteer recommendation.
more »
« less
Improving Efficiency of Volunteer-Based Food Rescue Operations
Food waste and food insecurity are two challenges that coexist in many communities. To mitigate the problem, food rescue platforms match excess food with the communities in need, and leverage external volunteers to transport the food. However, the external volunteers bring significant uncertainty to the food rescue operation. We work with a large food rescue organization to predict the uncertainty and furthermore to find ways to reduce the human dispatcher's workload and the redundant notifications sent to volunteers. We make two main contributions. (1) We train a stacking model which predicts whether a rescue will be claimed with high precision and AUC. This model can help the dispatcher better plan for backup options and alleviate their uncertainty. (2) We develop a data-driven optimization algorithm to compute the optimal intervention and notification scheme. The algorithm uses a novel counterfactual data generation approach and the branch and bound framework. Our result reduces the number of notifications and interventions required in the food rescue operation. We are working with the organization to deploy our results in the near future.
more »
« less
- Award ID(s):
- 1850477
- NSF-PAR ID:
- 10215658
- Date Published:
- Journal Name:
- Proceedings of the AAAI Conference on Artificial Intelligence
- Volume:
- 34
- Issue:
- 08
- ISSN:
- 2159-5399
- Page Range / eLocation ID:
- 13369 to 13375
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Hunger relief organizations are mostly non-profits that collect food from various sources and redirect them to the people in need. This is to combat the prevalent food insecurity affecting children, the unemployed, students, seniors and so on. Previous research has focused on the demand/donation side of food rescue operations, but the distribution or supply side - especially in reducing the uncertainty associated with food demand - has received significantly lower attention. In this study, we obtained data from a local hunger relief organization, specifically a food pantry to develop estimates of the demand they expect to receive in the future. To do this, we fit the growth of the food pantry client population to a logistic growth model to obtain a good fit. We then obtained data for frequency of visits to develop estimates of the number of visits expected in the future, using time series models. This will be combined with the allocation policy for food distribution to develop estimates of true demand. This study has merit for hunger relief organizations. It will aid decision making relative to food distribution, while also providing data for planning purposes.more » « less
-
null (Ed.)Reading and arithmetic are difficult cognitive feats for children to master and youth from low-income communities are often less “school ready” in terms of letter and number recognition skills ( Lee and Burkam, 2002 ). One way to prepare children for school is by encouraging caregivers to engage children in conversations about academically-relevant concepts by using numbers, recognizing shapes, and naming colors ( Levine et al., 2010 ; Fisher et al., 2013 ). Previous research shows that caregiver-child conversations about these topics rarely take place in everyday contexts ( Hassinger-Das et al., 2018 ), but interventions designed to encourage such conversations, like displaying signs in a grocery store, have resulted in significant increases in caregiver-child conversations ( Ridge et al., 2015 ; Hanner et al., 2019 ). We investigated whether a similar brief intervention could change caregiver-child conversations in an everyday context. We observed 212 families in a volunteer-run facility where people who are food-insecure can select food from available donations. Volunteers greet all the clients as they pass through the aisles, offer food, and restock the shelves as needed. About 25% of the clients have children with them and our data consist of observations of the caregiver-child conversations with 2- to 10-year-old children. Half of the observation days consisted of a baseline condition in which the quantity and quality of caregiver-child conversation was observed as the client went through aisles where no signs were displayed, and volunteers merely greeted the clients. The other half of the observation days consisted of a brief intervention where signs were displayed (signs-up condition), where, volunteers greeted the clients and pointed out that there were signs displayed to entertain the children if they were interested. In addition, there was a within-subject manipulation for the intervention condition where each family interacted with two different categories of signs. Half of the signs had academically-relevant content and the other half had non-academically-relevant content. The results demonstrate that the brief intervention used in the signs-up condition increases the quantity of conversation between a caregiver and child. In addition, signs with academically-relevant content increases the quality of the conversation. These findings provide further evidence that brief interventions in an everyday context can change the caregiver-child conversation. Specifically, signs with academically-relevant content may promote school readiness.more » « less
-
null (Ed.)Local business leaders, policy makers, elected officials, city planners, emergency managers, and private citizens are responsible for, and deeply affected by, the performance of critical supply chains and related infrastructures. At the center of critical supply chains is the food-energy-water nexus (FEW); a nexus that is key to a community’s wellbeing, resilience, and sustainability. In the 21st century, managing a local FEW nexus requires accurate data describing the function and structure of a community’s supply chains. However, data is not enough; we need data-informed conversation and technical and social capacity building among local stakeholders to utilize the data effectively. There are some resources available at the mesoscale and for food, energy, or water, but many communities lack the data and tools needed to understand connections and bridge the gaps between these scales and systems. As a result, we currently lack the capacity to manage these systems in small and medium sized communities where the vast majority of people, decisions, and problems reside. This study develops and validates a participatory citizen science process for FEW nexus capacity building and data-driven problem solving in small communities at the grassroots level. The FEWSION for Community Resilience (F4R) process applies a Public Participation in Scientific Research (PPSR) framework to map supply chain data for a community’s FEW nexus, to identify the social network that manages the nexus, and then to generate a data-informed conversation among stakeholders. F4R was piloted and co-developed with participants over a 2-year study, using a design-based research process to make evidence-based adjustments as needed. Results show that the F4R model was successful at improving volunteers’ awareness about nexus and supply chain issues, at creating a network of connections and communication with stakeholders across state, regional, and local organizations, and in facilitating data-informed discussion about improvements to the system. In this paper we describe the design and implementation of F4R and discuss four recommendations for the successful application of the F4R model in other communities: 1) embed opportunities for co-created PPSR, 2) build social capital, 3) integrate active learning strategies with user-friendly digital tools, and 4) adopt existing materials and structure.more » « less
-
Annotating medical images for the purposes of training computer vision models is an extremely laborious task that takes time and resources away from expert clinicians. Active learning (AL) is a machine learning paradigm that mitigates this problem by deliberately proposing data points that should be labeled in order to maximize model performance. We propose a novel AL algorithm for segmentation, ALGES, that utilizes gradient embeddings to effectively select laparoscopic images to be labeled by some external oracle while reducing annotation effort. Given any unlabeled image, our algorithm treats predicted segmentations as truth and computes gradients with respect to the model parameters of the last layer in a segmentation network. The norms of these per-pixel gradient vectors correspond to the magnitude of the induced change in model parameters and contain rich information about the model’s predictive uncertainty. Our algorithm then computes gradients embeddings in two ways, and we employ a center-finding algorithm with these embeddings to procure representative and diverse batches in each round of AL. An advantage of our approach is extensibility to any model architecture and differentiable loss scheme for semantic segmentation. We apply our approach to a public data set of laparoscopic cholecystectomy images and show that it outperforms current AL algorithms in selecting the most informative data points for improving the segmentation model. Our code is available at https://github.com/josaklil-ai/surg-active-learning.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1609/aaai.v34i08.7051
