A UK-based researcher of trustworthy explainable autonomous agency in multi-agent systems for achieving safer AI, with applications to autonomous vehicles.
Experience
Researching the intersection of AI safety and AI ethics through content analysis methods. Assistant to Dr. Atoosa Kasirzadeh of Carnegie Mellon University.
- Large scale quantitative literature analysis with unsupervised natural language processing tools.
- Curation, topic coding, and qualitative analysis of large corpora of papers.
Teaching assistant helping with the delivery and teaching of university courses.
- Assistant supervisor for two master's students.
- Online tutorial sessions to ~12 students for introductory machine learning course.
- Coursework and exam marker for courses in the School of Informatics, including Doing Research in NLP, Reinforcement Learning, Computer Systems, and Machine Learning.
Development and evaluation of novel planning and prediction algorithm for autonomous vehicles.
- Developed and evaluated IGP2, a goal-based interpretable prediction and planning system for autonomous vehicles with intuitive explanations.
- Publication at International Conference on Robotics and Automation (ICRA), 2021.
Projects
I am interested in combining technologies from explainable AI (XAI), causal reasoning, and natural language processing to support the creation of trustworthy and safe AI systems, with a focus on the domain of autonomous driving.
- CEMA: Combining counterfactual causation with RL-based planning to create causally-grounded explanations in naturual language
- Two large human subjects studies to elicit and evaluate naturally occuring and automatically generated explanations
- HEADD: The Human Explanations for Autonomous Driving Decisions dataset
- Integration of LLMs with CEMA in an RAG approach to improve the quality of explanations
- Curation of corpus of 3K+ papers on AI safety and AI ethics
- Data analysis and visualization techniques to identify overlapping and different topics in the literature
- Advocating for an epistemically inclusive approach to AI safety that considers long standing safe ML research and AI ethics
- Research done with Shannon Vallor and Atoosa Kasirzadeh
Developed methods to improve the interpretability and verifiability of autonomous vehicle decision-making.
- IGP2: Implementing rational inverse planning and Monte Carlo Tree Search for interpretable goal-based prediction and planning in autonomous vehicles
- GRIT: Training and evaluating decision tree-based verifiable goal recognition models for autonomous driving
Education
Publications
Balint Gyevnar and Cheng Wang and Christopher G. Lucas and Shay B. Cohen and Stefano V. Albrecht
Anton Kuznietsov* and Balint Gyevnar* and Cheng Wang and Steven Peters and Stefano V. Albrecht
Balint Gyevnar
Balint Gyevnar and Stephanie Droop and Tadeg Quillien and Shay B. Cohen and Neil R. Bramley and Christopher G. Lucas and Stefano V. Albrecht
Balint Gyevnar and Nick Ferguson and Burkhard Schafer
Balint Gyevnar and Massimiliano Tamborski and Cheng Wang and Christopher G. Lucas and Shay B. Cohen and Stefano V. Albrecht
Cillian Brewitt and Balint Gyevnar and Samuel Garcin and Stefano V. Albrecht
Stefano V. Albrecht and Cillian Brewitt and John Wilhelm and Balint Gyevnar and Francisco Eiras and Mihai Dobre and Subramanian Ramamoorthy
Awards
Colours reward those individuals have given time and effort above and beyond the call of duty to their chosen sport or Club. University Sport could not operate without these volunteers organising and co-ordinating Clubs and the Colours award recognises this endeavour.
Researchers from 18 countries answered the call, offering intriguing perspectives on AI and its impacts on society. In addition to the winner, AI100 selected a collection of five essays that thoughtfully consider AI at the intersection of morality, regulation, love, labor, and religion. Balint's essay was selected as one of the five finalists, titled 'Love, Sex, and AI'.
The inaugural TAS Early Career Researcher (ECR) Awards were initiated to celebrate outstanding contributions made by PhD students and postdoctoral researchers to any area of Trustworthy Autonomous Systems research in the last three years. Balint was awarded £4000 by the UKRI TAS Hub to achieve his vision for more trustworthy autonomous systems (TAS) through explainability and conversations.
In 2021, the Intelligent Transportation Systems Society (ITSS) organized a competition on the topic “Shape the Future of ITS” by asking participants to present a futuristic vision on transportation systems, the way they will operate, and the product and services they will provide. Balint Gyevnar's vision was selected as a 3rd place winner, titled 'Cars that Explain: Building Trust in Autonomous Vehicles through Explanations and Conversations'.
Volunteer
Executive committee member of EUVC responsible for the deliverance of the club's volleyball programme to more than 200 members.
- Responsible for public outreach to and networking with alumni members and organizing a two-day event series.
- Large-scale event organization, public speaking, timetabling, and human resource management of 8 teams, 10 coaches, and 220+ members.
- Managing a cash flow of approximately £70k, setting up an annual budget, and managing thousands of transactions.