Bridging Neural and Symbolic Representations with Transitional Dictionary Learning

Authors: Junyan Cheng, Peter Chin

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments are conducted on three abstract compositional visual object datasets, which require the model to utilize the compositionality of data instead of simply exploiting visual features. Then, three tasks on symbol grounding to predefined classes of parts and relations, as well as transfer learning to unseen classes, followed by a human evaluation, were carried out on these datasets.
Researcher Affiliation Academia Junyan Cheng Thayer School of Engineering Dartmouth College Hanover, NH 03755, USA jc.th@dartmouth.edu Peter Chin Thayer School of Engineering Dartmouth College Hanover, NH 03755, USA pc@dartmouth.edu
Pseudocode Yes Algorithm 1 Decompose an input
Open Source Code Yes Our code and data are available at https://github.com/chengjunyan1/TDL.
Open Datasets Yes Omni Glot (Lake et al., 2015) contains handwritten characters. Shape Net5 is composed of 3D shapes in 5 categories (bed, chair, table, sofa, lamp) from Shape Net (Chang et al., 2015) voxelized by binvox (Min, 2004 2019).
Dataset Splits Yes In total, we synthesize 50000 samples, which are divided into 8:1:1 splits for training, development, and testing.
Hardware Specification Yes We conducted our experiments on our internal clusters, and a major workload has the following configuration: six Quadro RTX 5000 GPUs and one Quadro RTX 8000 GPU, along with an Intel (R) Xeon (R) Silver 4214R CPU @ 2.40GHz and 386 GB RAM.
Software Dependencies No The paper mentions software like PyTorch Lightning and Weights & Biases Sweep but does not provide specific version numbers for these or other key software components.
Experiment Setup Yes Table 3: Hyper-parameter search distributions. lr {1e 3, 2e 3, 5e 4} th S U(0.1, 0.5; 0.05) or None K U(3, 9; 1) quota U(8, 32; 4) αoverlap U(0.1, 0.25) demb {64,128,256,512} αresources U(0.05, 0.2) dsampler {32,64} γcluster U(5e 3, 2e 2) dmapper {128,256,512} σ {2.5,5,10,15,25}