Imitation of Life: A Search Engine for Biologically Inspired Design
Authors: Hen Emuna, Nadav Borenstein, Xin Qian, Hyeonsu Kang, Joel Chan, Aniket Kittur, Dafna Shahaf
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments demonstrate that BARCODE can retrieve inspirations that are valuable to engineers and designers tackling real-world problems, as well as recover famous historical BID examples. [...] In this section, we report our evaluation of the performance of BARCODE s bio-inspiration score module, as well as an end-to-end evaluation of the system s ability to retrieve biomimetic inspirations. |
| Researcher Affiliation | Academia | Hen Emuna1, Nadav Borenstein2, Xin Qian3, Hyeonsu Kang4, Joel Chan3, Aniket Kittur4, Dafna Shahaf1 1The Hebrew University of Jerusalem 2University of Copenhagen 3University of Maryland 4Carnegie Mellon University |
| Pseudocode | No | The paper describes its algorithms and processes in textual format and through flowcharts (e.g., Figure 1), but it does not contain explicit pseudocode or algorithm blocks. |
| Open Source Code | Yes | We release data and code at https://github.com/emunatool/BARcode Bio Inspired-Search. |
| Open Datasets | Yes | BARCODE can work with any corpus containing biological knowledge. For this paper, we focus on Wikipedia because of its accessibility and broad coverage of diverse biological topics (Wikimedia.org 2019). The corpus contains 780,949 sentences, collected from 27,640 articles listed under the category Articles with species microformats 1. [...] We used a dataset of patents to construct a list of known problems (using what one might call cross-corpus distant supervision ): First, we downloaded 20 million sentences from the Claims section of a random set of patents. |
| Dataset Splits | No | The paper describes a filtering process resulting in '23,553 sentences (3% of the total data)' and mentions training a classifier, but it does not provide explicit train/validation/test splits for the main experimental datasets used (Wikipedia corpus, patents) or the crowd-sourced evaluation data. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU models, CPU models, memory specifications) used for conducting the experiments. |
| Software Dependencies | Yes | We used spa Cy s NLP engine (Honnibal and Montani 2017) to process the articles. [...] We employed SBERT, designed for semantic search (Reimers and Gurevych 2019)5, and De BERTa (He et al. 2020)6 as the inference model. [...] https://huggingface.co/sentence-transformers/multi-qampnet-base-dot-v1 6https://huggingface.co/cross-encoder/nli-deberta-v3-base |
| Experiment Setup | No | The paper describes the system architecture and evaluation process, but it does not provide specific experimental setup details such as hyperparameter values (e.g., learning rates, batch sizes, number of epochs) or other fine-grained training configurations for the models used (SBERT, DeBERTa, simple classifier). |