Primary supervisor
Ehsan ShareghiLLMs have enabled us to break the performance ceiling in several tasks. Language Agents took this further by connecting language models with tools, environments and enriched them with memory and feedback. Could this idea be applied to automating a machine learning task from the start (i.e., description of a problem) all the way to designing the machine learning methodology (i.e., algorithm and method design), implementation (i.e., implementing the model in PyTorch), evaluating its loss during training (i.e., to detect if the design was flawed or well-behaving), and post analysis and further refinement (i.e., analysing the learning curves and evaluation metrics to further improve the design)?
Student cohort
Double Semester
Aim/outline
- An open source code
- A small high quality dataset
- A publication in ACL/EMNLP/NAACL/EACL
URLs/references
- AutoML https://www.automl.org/automl/
- Large Language Model Agent for Hyper-Parameter Optimization
- Automatically Generating Deployable Models from Language Instructions for Computer Vision Tasks
- https://openai.com/index/mle-bench/
Required knowledge
- Must: fluency in Python and PyTorch
- Must: academic or Working knowledge of Large Language Models
- Must: fluent in basic machine learning concepts (both theory and hands-on)
- Preferred: have built a small fine-tuned language model (i.e., LLaMA)
- Preferred: prior experience with language agents