Eoin Kenny
Nominated Award: Best Application of AI In a Student Project
Website of Company: https://test.insight-centre.org/users/eoin-kenny
University College Dublin (the company) is an educational institution; specifically, the nominee works across two research institutes in the School of Computer Science: the Insight SFI Centre for Data Analytics and the VistaMilk SFI Research Centre. Insight is one of the largest AI centres of its kind in Europe, with over 400 researchers, €100 million in funding, over 80 industry partners, and 8 research institutions. VistaMilk is a SFI and DAFM funded research centre directed at the one of Ireland’s largest indigenous sectors, the Agri-Food Diary sector. The purpose of the centre is to combine Agriculture and ICT expertise to innovate and enhance sustainability across the dairy supply chain.
The work was motivated by two considerations:
1) The first 15 years of AI-R&D funded by SFI in Ireland mainly benefited the US/E U multinational technology sector (e.g.. Microsoft, Intel, Ericsson, IBM, Siemens) through partnerships with Irish Universities. The indigenous Irish multinational, SME, and start-up sectors hardly benefited at all. Hence, a guiding motivation for this work was to benefit Ireland’s indigenous sectors, specifically the Agri-Food sector, to provide advanced technology research and development support in AI to smaller firms (e.g., farmers and growers) and larger indigenous multinationals (e.g., food companies). 2) The work was also motivated by the enormous challenges of climate change and sustainability impacting the farming sector; specifically, those facing dairy farmers and the Agri-Food sector as a whole.
The aim of this research was to produce an AI solution that would be a working system designed to be widely-used by dairy farmers in Ireland. The aforementioned system should help farmers better manage their farms from a sustain ably perspective by providing accurate grass-growth prediction for feed bud geting for their herds. Accurate forage prediction and budgeting underlies better farm management as it can help reduce fertiliser use (reducing nitrate pollution) and the use of carbon-intensive feed supplements (replacing them with on-farm grass that has much lower transport costs). PastureBase Ireland (PBI, www.pbi.ie) is a decision support system for dairy-farm pasture management, delivered by Teagasc to several thousand dairy farmers. The long-term aim of this work is to develop an accurate predictive system combining first- principles mechanical models (developed by Teagasc) paired with AI models that use historical data (developed by UCD) to solve the grass- growth prediction problem. This AI system
also tackles the accompanying problem of explaining the model’s predictions (ie, explainable AI), as well as trying to handle the difficulties that arise in predicting in the context of climate change.
Reason for Nomination:
We nominate the work of Eoin Kenny (PhD student at UCD) to recognise the innovative and bleeding-edge technological breakthroughs he has made in the last three years in predictive AI modelling, explainable AI and precision agriculture. These discoveries have been recognised internationally by the AI community at major AI conferences and through a Best Paper Award. It would be nice for him to be recognised in his home-country too.
His contributions have technological and scientific impacts, but also have significant societal and economic impact for the future of the Irish Agri-Food sector, specifically in the context of sustain ability and climate change.
TECHNOLOGICAL & SCIENTIFIC IMPACTS
Kenny’s AI work involved solving several key problems using state-of-the-art techniques that have been recognised internationally as ground-breaking and highly significant (see Additional Information for details).
Three major AI solutions were developed to solve problems in crop-growth prediction for the dairy sector: (i) the accurate prediction of grass growth from historical data, (ii) the automated explanation of the model predictions to farmer end- users, (iii) the prediction of grass growth under climate disruption, when historical data becomes less useful. Consider each in turn:
PREDICTIVE AI FOR GRASS GROWTH
The Teagasc PastureBase Ireland system holds a large dataset of on-farm reports of grass growth entered by end-users (farmers) throughout the year for many years, along with environmental conditions. The problem was to build a predictive A I model using this historical data. However, though extensive, this dataset of almost 100,000 records was very noisy (e.g., including miss-entries and partial records). The AI model Kenny developed, called PBI-CBR, achieve d the required accuracy predicting grass growth (a week ahead) for individual farms using a case-based reasoning (CBR) or k-NN model. The model’s main novelty was a Bayesian case-exclusion method that cleaned the noisy dataset; it removes records that fall significantly outside the distribution of a gold-standard research dataset (compiled by Teagasc ’s in Fermoy). This AI data-cleaning method won a best paper award at ICCBR-19 (Osnabrück, Germany).
EXPLAINABLE AI
On top of this innovative predictive model, Kenny also developed an eXplainable AI (XAI) module to explain the model’s predictions. This module used example-based explanations, finding very similar nearest-neighbouring casesto a target case for a given week; specifically, the system found explanatory cases for the same-farm as the target-case telling the farmer that “The model predicts grass growth of 10kg/ha/DM next week because it looks like a week two-years ago on your farm with very similar climate and environmental conditions. Example-based explanation in XAI was well-known, but Kenny’s innovation has been to extend this explanation strategy to Deep Learning models too (by abstracting feature-weights from the neural network). He has also innovated further developing new example-based strategies involving counterfactuals and semi-factuals. This XAI work has been published at the A* leading international AI conferences in the last three years (see IJCAI-19, IJCAI-20, IJCAI-21, AAAI-21) giving Kenny an unprecedented and uniquely successful research profile ahead of any current AI researcher in Ireland.
COMBINED HYBRID MODELLING
Finally, Kenny has combined AI models of grass-growth prediction (using a modified neural network model) with traditional mechanical models to gain the best of both modelling worlds to handle the vagaries of climate change. Teagasc traditionally uses a mechanical model (called MoSt that formulaically relates climate variables to soil and other chemical factors affecting grass growth). Data analyses revealed that the AI model worked better during climate disruptive events (e.g., the 2018 drought and forage crisis), but the mechanical model works better during more-normal periods. So, Kenny built an AI mode l that learns which model works best for a given time-period, selecting the best model to use for different prediction periods. Therefore, this hybrid model is better positioned to deal with future climate change.
SOCIETAL & ECONOMIC IMPACTS
These AI modelling efforts positively impact the everyday lives of people immediately. It offers solutions to the problem of grass growth prediction for a major indigenous sector, offering this sector a technological basis for a more sustainable agriculture, one that better manages a sustainable resource (grass), while minimising the potentially polluting effects of farming activities. The path to use of these innovations is also clear, as they can be integrated into the existing PastureBase Ireland platform to provide direct predictive capabilities to several 1000 dairy farmers. This state-of-the-art AI work in AI promises to improve the international competitiveness of this Irish Agri-Food sector.
Additional Information:
The nominee, Eoin Kenny, has a uniquely successful publication/award profile amongst the 100s of AI researchers in Ireland over the last few years, with over 15 publications for
his PhD work (see https://scholar.google.com/citations?user=AzMTFY4AAAAJ&hl=en ).
These publications are in some of the most prestigious venues for AI research, including.
1) Kenny, Eoin M., and Mark T. Keane. “Twin-systems to explain artificial neural networks using case-based reasoning: Comparative tests of feature-weighting methods in ANN-CBR twins for XAI.” IJCAI-19, Macao, 10-16 August 201 9. 2019. A* Core conference, with 80%+ Rejection Rate. Ranked 9th in Google-Scholar Global Impact Rankings
2) Kenny, E. M., Ruelle, E., Geoghegan, A., Shalloo, L., O’Leary, M., O ’Donovan, M., & Keane, M. T. (2019, September). Predicting grass growth for sustainable dairy farming: A CBR system using bayesian case-exclusion and post-hoc, personalized explanation-by-example (XAI). In International Conference on Case-Based Reasoning (pp. 172-187). Springer, Ch am. B Core conference, with 60%+ Rejection Rate. Best Paper Award
3) Kenny, E. M., Ruelle, E., Geoghegan, A., S halloo, L., O’Leary, M., O’Donovan, M., & Keane, M. T. (202 0). Bayesian Case- Exclusion and Explainable AI (XAI) for Sustainable Farming. IJCAI-20, Montreal, 2020 (Best Paper Track) A* Core conference, with 80%+ Rejection Rate
Ranked 9th in Google-Scholar Global Impact Rankings
4) Kenny, E. M., & Keane, M. T. (2021, May). On Generating Plausible Counterfactual and Semi-Factual Explanations for Deep Learning. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 35, No. 13, pp. 11575-11585). A* Co re conference, with 80%+ Rejection Rate Ranked 4th in Google -Scholar Global Impact Rankings
5) Kenny, E. M., Ford, C., Quinn, M., & Keane, M. T. (2021). Explaining black-box classifiers using post-hoc explanations-by-example: The effect of explanations and error-rates in XAI user studies. Artificial Intelligence, 294, 103459. Journal Impact Factor 9.088+
6) Kenny, E. M., & Keane, M. T. (202 1, Dec). Explaining Deep Learning Using Examples: Optimal Feature Weighting Methods for Twin Systems Using Post-Hoc, Explanation-by-Example in XAI. Knowledge Based Systems Journal Impact Fact or 8.038 Ranked 13th in Google-Scholar Global Impact Rankings