Max-margin Latent Dirichlet Allocation for Image Classification and Annotation

By Yang Wang, Greg Mori. 2011

38 citations recorded by the source corpus.

We present the max-margin latent Dirichlet allocation, a max-margin variant of supervised topic models, for image classification and annotation. Our model for image classification (called MMLDA c ) integrates discriminative classification with generative topic models. Our model for image annotation (called MMLDA a ) extends MMLDA c to the case of multi-label problems, where each image can be associated with more than one annotation terms. We derive efficient learning algorithms for both models and demonstrate experimentally the advantages of our proposed models over other baseline methods.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Probabilistic & Bayesian Methods

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Eric P. Xing
2. Jun Zhu
3. Yin Zheng
4. Yoshua Bengio
5. Yujin Zhang
6. Fan Hu
7. Hugo Larochelle
8. Amr Ahmed

In Computer Vision and Pattern Recognition

1. Eric P. Xing
2. Jun Zhu
3. Fan Hu
4. Yin Zheng
5. Ming-Hsuan Yang
6. Luc Van Gool
7. Yoshua Bengio
8. Junwei Han

In Probabilistic & Bayesian Methods

1. Yin Zheng
2. Yujin Zhang
3. Jun Zhu
4. Eric P. Xing
5. Amr Ahmed
6. Fan Hu
7. Hong Sun
8. Hamed Jelodar

Citations in the corpus

• Self-Supervised Learning of Visual Features through Embedding Images into Text Topic Spaces (2017, neutral): “Supervised variations of LDA are presented in [30, 42, 28] where the discovered topics are driven by the semantic regularities of interest for the classification task.”
• Latent Dirichlet allocation (LDA) and topic modeling: models, applications, a survey (2017, methodology): “According to the Table 1, used LDA model for variety subjects, such as: Scientific topic discovery [35], Source code analysis [27], Opinion Mining [30], Event detection [40], Image Classification [41].”
• A Deep and Autoregressive Approach for Topic Modeling of Multimodal Data (2014, methodology): “Wang et al. [17] opted instead for a maximum margin formulation of LDA (MMLDA).”
• Tile-Level Annotation of Satellite Images Using Multi-Level Max-Margin Discriminative Random Field (2013, background): “The most related work to ours is detailed in [16], where the authors only exploit different types of feature representation and do not make full use of the contextual information that may be beneficial to annotation tasks.”
• Topic Modeling of Multimodal Data: An Autoregressive Approach (2014, methodology): “LabelMe UIUC-Sports Model Accuracy F-measure Accuracy F-measure SPM [17] 80.88% 43.68% 72.33% 41.78% MMLDA [12] 81.47%† 46.64%†∗ 74.65%† 44.51%† sLDA [9] 81.87% 38.7%† 76.87% 35.0%† DocNADE 81.97% 43.32% 74.23% 46.38% SupDocNADE 83.43% 43.87% 77.29% 46.95% †: Taken from the original paper.”
• MedLDA: maximum margin supervised topic models (2012, background): “As shown in the above experiments, our simulation experiments below, as well as the follow-up works (Yang et al., 2010; Wang and Mori, 2011; Li et al., 2011), depending on the data and problems, max-margin supervised topic models can outperform SVM models, or they are comparable if no gains on predi”