Multi-instance multi-label learning in the presence of novel class instances
Authors: Anh Pham, Raviv Raich, Xiaoli Fern, Jesús Pérez Arriaga
ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on both synthetic and real datasets illustrate the effectiveness of the proposed approach. In this section, we compare our approach with related methods in MIML learning using MIML data that contains novel class instances. |
| Researcher Affiliation | Academia | Anh T. Pham PHAMAN@EECS.OREGONSTATE.EDU Raviv Raich RAICH@EECS.OREGONSTATE.EDU Xiaoli Z. Fern XFERN@EECS.OREGONSTATE.EDU School of Electrical Engineering and Computer Science, Corvallis, OR 97330-5501 USA / Jes us P erez Arriaga JPEREZ@GTAS.DICOM.UNICAN.ES Departamento de Ingenier ıa de Comunicaciones, Universidad de Cantabria, 39005 Santander, Spain |
| Pseudocode | Yes | Pseudo code for computing the probability p(ybi = c, Yb = L|Xb, w) is provided in algorithm 1 in the supplementary material. |
| Open Source Code | No | No explicit statement or link is provided for the public release of the source code for the methodology described in this paper. |
| Open Datasets | Yes | We use MSCV2, Letter Carroll, and Letter Frost datasets (Briggs et al., 2012), MNIST handwritten dataset (Asuncion & Newman) in these experiments. |
| Dataset Splits | No | The paper describes how datasets are constructed (e.g., generating 100 bags for MNIST, splitting classes into known/unknown for MSCV2, Letter Carroll, and Letter Frost), but does not provide explicit percentages, counts, or references to standard train/validation/test splits for reproducibility. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, memory amounts, or cloud instances) are provided for the experimental setup. |
| Software Dependencies | No | The paper mentions using specific techniques and referencing other methods, but it does not provide specific software names with version numbers (e.g., 'Python 3.x', 'PyTorch 1.x') for reproducibility. |
| Experiment Setup | Yes | The parameter λMIML NC for the kernel version of the proposed approach is fixed at 10 5. The parameter λKS for the kernel scoring approach is searched over {10 4, 10 2, 10 1, 100}. and We apply gradient ascent with backtracking line search to maximize g(w, w ) w.r.t. w... and An unlabeled test instance is detected as novel instance if p(yti = 0|xti, w) θ, where 0 θ 1 is a manually selected threshold. |