On the Optimal Bit Complexity of Circulant Binary Embedding
Authors: Saehoon Kim, Jungtaek Kim, Seungjin Choi
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We also provide numerical experiments to support our theoretical results. In this section, we conducted various numerical experiments to support the theoretical analysis of CBE developed in section with the following datasets: MNIST..., CIFAR-10..., GIST1M... Figure 3: Plots for the relative error on approximating the angle between vectors measured by Frobenious and spectral norms, showing that CBE is almost identical to the standard binary embedding. |
| Researcher Affiliation | Collaboration | Saehoon Kim AItrics, Korea and Department of Computer Science and Engineering Pohang University of Science and Technology 77 Cheongam-ro, Nam-gu, Pohang 37673, Korea kshkawa@postech.ac.kr Jungtaek Kim, Seungjin Choi Department of Computer Science and Engineering Pohang University of Science and Technology 77 Cheongam-ro, Nam-gu, Pohang 37673, Korea {jtkim,seungjin}@postech.ac.kr |
| Pseudocode | No | h C(x) = sgn F 1(F(g) F(Dx) , (7) where F( ) represents the discrete Fourier transform, F 1( ) is the inverse discrete Fourier transform, and is element-wise product. This is a mathematical description, not pseudocode. |
| Open Source Code | No | No statement regarding the release of source code for the methodology is found in the paper. |
| Open Datasets | Yes | MNIST (Le Cun et al. 1998) consists of 70,000 handwritten digit images where images are represented by 784-dimensional vectors. CIFAR-10 (Krizhevsky and Hinton 2009) consists of 60,0000 low-resolution images from 10 classes. GIST1M (J egou, Douze, and Schmid 2011) consists of 920-dimensional 1 million GIST descriptors with additional 1,000 queries. |
| Dataset Splits | No | To evaluate H and A, we randomly select 3,000 data points for all datasets. This describes a subset used for evaluation, not a full training/validation/test split for reproducibility of model training. |
| Hardware Specification | No | No specific hardware details such as GPU/CPU models, memory, or processor types are mentioned in the paper. |
| Software Dependencies | No | No specific software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9) are mentioned in the paper. |
| Experiment Setup | No | All experiments are repeated five times to avoid any bias. For preprocessing, all vectors in datasets are l2 normalized. These are general experimental practices, not specific hyperparameters or detailed training configurations for the main CBE method. |