Demystifying Embedding Spaces using Large Language Models
Authors: Guy Tennenholtz, Yinlam Chow, ChihWei Hsu, Jihwan Jeong, Lior Shani, Azamat Tulepbergenov, Deepak Ramachandran, Martin Mladenov, Craig Boutilier
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate our approach on a variety of diverse tasks, including: enhancing concept activation vectors (CAVs), communicating novel embedded entities, and decoding user preferences in recommender systems. Our work couples the immense information potential of embeddings with the interpretative power of LLMs. |
| Researcher Affiliation | Industry | Guy Tennenholtz , Yinlam Chow, Chih-Wei Hsu, Jihwan Jeong, Lior Shani, Azamat Tulepbergenov, Deepak Ramachandran, Martin Mladenov, Craig Boutilier Google Research |
| Pseudocode | No | The paper does not include any sections or figures explicitly labeled 'Pseudocode', 'Algorithm', or similar to present structured steps in an algorithm block format. |
| Open Source Code | No | The paper does not contain an explicit statement about releasing its source code or provide any link to a code repository for the methodology described. |
| Open Datasets | Yes | We use the Movie Lens 25M dataset (Harper & Konstan, 2015), which we enrich with textual descriptions by generating a large corpus of text using a Pa LM 2-L (Unicorn) LLM (Google et al., 2023). |
| Dataset Splits | No | For the movie tasks, we use 1000 randomly sampled examples for test and the rest for training. For the user profile task we use an 80/20 training/test split. The paper mentions train and test splits, but does not explicitly specify a validation split or its size/methodology for reproducibility. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for its experiments, such as particular GPU or CPU models, or cloud computing instance types. |
| Software Dependencies | No | The paper mentions fine-tuning models 'using a pretrained Pa LM 2-XS (Otter)' but does not provide specific version numbers for software libraries or environments like Python, PyTorch, or TensorFlow. |
| Experiment Setup | Yes | Particularly, for movie tasks, we run the first stage of training for 20k iterations, and then the second stage for another 300k iterations, using a batch size of 32 (i.e., roughly seven epochs). |