Universal In-Context Approximation By Prompting Fully Recurrent Models
Authors: Aleksandar Petrov, Tom Lamb, Alasdair Paren, Philip Torr, Adel Bibi
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | There are two experimental aspects to this work. First, there is the implementation of LSRL and the two universal approximation programs in Lsts. 1 and 2. The most critical aspect of implementing LSRL is the debranching algorithm which is described in detail in App. A. Additionally, the two programs are described in full in their corresponding listings. We also provide Python implementation for the LSRL compiler and the two programs. Second, there is the study of how affected by parameter noise are the different implementations of the conditional assignment operator f_ifelse which was presented in Sec. 6. The details of this experiment are described in Fig. 4 and we also provide the code with which we did the experiment and our plots. |
| Researcher Affiliation | Academia | Aleksandar Petrov, Tom A. Lamb, Alasdair Paren, Philip H.S. Torr, Adel Bibi Department of Engineering Science University of Oxford |
| Pseudocode | Yes | Listing 1: LSRL program for universal approximation in-context for continuous functions. |
| Open Source Code | Yes | Our implementation of LSRL is available at https://github.com/Aleksandar Petrov/LSRL |
| Open Datasets | No | The paper does not use or refer to any publicly available or open datasets for training or evaluation. The experiments are based on constructed models. |
| Dataset Splits | No | The paper's experiments are based on constructed models and do not involve dataset splits for training, validation, or testing. |
| Hardware Specification | No | The paper mentions that experiments were run using 'only CPU compute' but does not provide specific details such as CPU models, memory, or other hardware specifications. |
| Software Dependencies | No | The paper mentions 'Sci Py.sparse, Virtanen et al. 2020' and 'Sym Py, Meurer et al. 2017' as software used, but does not specify exact version numbers for these or other key software components. |
| Experiment Setup | No | The paper's experiments are based on constructed models and theoretical analysis, thus do not involve traditional training setups with hyperparameters, optimizers, or batch sizes. |