Ekin Akyürek
reinforcement learning
robustness
analysis
evaluation
generalization
compositional generalization
compositionality
recurrent neural networks
generalizability
semantic parsing
memorization
rnn
sequence-to-sequence
systematic generalization
natural language feedback
3
presentations
SHORT BIO
Ekin Akyürek is a rising third year PhD student in Computer Science at MIT. He is working with Prof. Jacob Andreas on compositional generalization at MIT-CSAIL. His recent research focuses on building neural network models with inductive biases that support human-like generalization (e.g. to support compositionality and one-shot word learning). Previously, he worked with Prof. Alan Edelman on automatic differentiation, and with Prof. Deniz Yuret on sequence-to-sequence modelling / morphological analyses. He received his Bachelor’s degrees in Electrical & Electronics Engineering and in Physics from Koç University in 2019.
Presentations
Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks
Zhaofeng Wu and 8 other authors
RL4F: Generating Natural Language Feedback with Reinforcement Learning for Repairing Model Outputs
Afra Feyza Akyurek and 5 other authors
Lexicon Learning for Few Shot Sequence Modeling
Ekin Akyürek and 1 other author