Kernelized Bayesian Transfer Learning
Authors: Mehmet Gönen, Adam Margolin
AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We illustrate the generalization performance of our algorithms on two different applications. In computer vision experiments, our method outperforms the state-of-the-art algorithms on nine out of 12 benchmark supervised domain adaptation experiments defined on two object recognition data sets. In cancer biology experiments, we use our algorithm to predict mutation status of important cancer genes from gene expression profiles using two distinct cancer populations, namely, patient-derived primary tumor data and in-vitro-derived cancer cell line data. [...] 3 Experiments We first test our new algorithm KBTL on 12 benchmark domain adaptation experiments derived from two computer vision data sets to illustrate its generalization performance and compare its results to previously reported results on these experiments. We then perform transfer learning experiments with heterogeneous populations for two different cancer types to show the suitability of our algorithm in a challenging and nonstandard application scenario. |
| Researcher Affiliation | Collaboration | Mehmet G onen mehmet.gonen@sagebase.org Sage Bionetworks Seattle, WA 98109, USA Adam A. Margolin adam.margolin@sagebase.org Sage Bionetworks Seattle, WA 98109, USA Present address: Department of Biomedical Engineering, Oregon Health & Science University, Portland, OR 97239, USA. |
| Pseudocode | No | The paper contains graphical models and mathematical equations but no pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our Matlab implementations for binary and multiclass classification are available at https://github.com/mehmetgonen/kbtl. |
| Open Datasets | Yes | In this first set of experiments, we use Office (Saenko et al. 2010) and Caltech-256 (Griffin, Holub, and Perona 2007) data sets. [...] We use primary tumor data from The Cancer Genome Atlas (TCGA) (TCGA Research Network 2008) and cancer cell line data from the Cancer Cell Line Encyclopedia (CCLE) (Barretina et al. 2012). |
| Dataset Splits | Yes | To have results comparable to the previous studies, we follow the experimental setup used by Saenko et al. (2010), Gong et al. (2012), and Hoffman et al. (2013), and perform experiments for 20 random train/test splits provided by Hoffman et al. (2013). [...] For each replication, we randomly select 25 per cent of TCGA with stratification as the test set and use 25, 50, or 75 per cent as the training set, whereas we use all data points in CCLE for training. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) are mentioned for the experimental setup. |
| Software Dependencies | No | The paper states 'Our Matlab implementations' but does not specify a Matlab version or any other software dependencies with version numbers. |
| Experiment Setup | Yes | For our algorithm, the hyper-parameter values are selected as ( , β ) = ( γ, βγ) = ( λ, βλ) = (1, 1), σh = 0.1, and = 1. The number of components in the hidden representation space is selected as R = 20. We take 200 iterations for variational inference scheme. [...] For our algorithm, the number of components in the hidden representation space is selected as R = 2. |