TRAILS AI Summer Academy at the University of Maryland
TRAILS AI Summer Academy at the University of Maryland
Embedded Files
Summer 2023 Research Presentations
Summer 2023 Research Presentations
For more pictures of how much fun camp was this summer, click here!
AI4ALL SUMMER 2023
A two-week summer camp, exposing students to personal growth, educational, and hands-on experiences in artificial intelligence presented by faculty, guest lecturers, and University of Maryland students.
Meet our Participants
Project Groups
Project Groups
Project 1: Attack
This project runs adversarial attacks on image processing algorithms, which are "malicious attacks on the data which may seem okay to a human eye but causes misclassification in a machine learning pipeline". Through these adversarial attacks, students can study the behavior of normal and adversarially robust neural networks and observe the corresponding output, which gives insight into what parts of the input are relevant to the neural net. During the progression of this project, students trained image classifiers, learned and implemented adversarial attacks on neural networks, and used attacks to interpret features of models.Project 1: Attack
Project 2: Visual
This project focuses on helping students visualize and explain a machine-learning model's prediction or answer. By pushing students to think about how these models make their decisions, students started to wonder what features were most important in making this decision. By implementing machine learning models and applying explainability techniques such as LIME and SHAP, students were able to quantify the contribution of different features to the model's decision.Project 2: Visual
Project 3: Learn
Reinforcement learning is rising in its significance in our everyday world. From Teslas and ChatGPT to video games, the benefits and applications of reinforcement learning are evident. Students were prompted to think about reinforcement learning and how they could apply its basics to programming an online "test" car. Depending on the environment, the goal for their car was different, but throughout the process, they were challenged to see if altering their programs' reward systems could improve their cars' abilities to perform well.Project 3: Learn
Project 4: Crowds
When moving towards a goal, robots rely on explicit communication or instructions to avoid collisions or getting stuck. The ability to model these interactions can help robots predict the uncertain behavior of pedestrians in the absence of explicit communication. This project developed a model of crowd interactions that could allow robots to safely navigate to their destination.Project 4: Crowds
Project 5: Riddle
In this project, students explored how AI systems can be used for generating and answering riddles, as a specific example of the broader question of how AI systems are built in general. Students tested AI’s understanding of a fun riddles data set, interactively playing with a chatbot such as ChatGPT and training their own models for answering riddles. The students learned how to prompt large language models to perform tasks, how to use crowdsourcing to collect human judgments, and how to train their own natural language processing models to answer riddles.Project 5: Riddle
Page updated
Report abuse