Anh T. Pham

Efficient instance annotation in multi-instance learning

The cost associated with manually labeling every individual instance in large datasets is prohibitive. Significant labeling efforts can be saved by assigning a collective label to a group of instances (a bag). This setup prompts the need for algorithms that allow labeling individual instances (instance annotation) based on bag-level labels. Probabilistic models in which instance-level labels are latent variables can be used for instance annotation. Brute-force computation of instance-level label probabilities is exponential in the number of instances per bag due to marginalization over all possible combinations. Existing solutions for addressing this issue include approximate methods such as sampling or variational inference. This paper proposes a discriminative probability model and an expectation maximization procedure for inference to address the instance annotation problem. A key contribution is a dynamic programming solution for exact computation of instance probabilities in quadratic time. Experiments on bird song, image annotation, and two synthetic datasets show a significant accuracy improvement by 4%-14% over a recent state-of-the-art rank loss SIM method.

Supervised hierarchical segmentation for bird song recording

A common framework of identifying bird species from audio recordings involves detecting bird song segments, which will be subsequently input to a classifier. In-field recordings are contaminated with various environmental noise. For such recordings, supervised segmentation has been observed to outperform unsupervised energy-based approaches. Prior supervised segmentation work considers only pixel-level predictions and ignores the supervision provided at the segment-level. We propose a hierarchical approach that learns to isolate bird song syllables based on both pixel-level and segment-level information. Experimental results suggest that our method outperforms an existing supervised method that learns only from pixel-level supervision.

Dynamic Programming for Instance Annotation in Multi-Instance Multi-Label Learning

Labeling data for classification requires significant human effort. To reduce labeling cost, instead of labeling every instance, a group of instances (bag) is labeled by a single bag label. Computer algorithms are then used to infer the label for each instance in a bag, a process referred to as instance annotation. This task is challenging due to the ambiguity regarding the instance labels. We propose a discriminative probabilistic model for the instance annotation problem and introduce an expectation maximization framework for inference, based on the maximum likelihood approach. For many probabilistic approaches, brute-force computation of the instance label posterior probability given its bag label is exponential in the number of instances in the bag. Our contribution is a dynamic programming method for computing the posterior that is linear in the number of instances. We evaluate our method using both benchmark and real world data sets, in the domain of bird song, image annotation, and activity recognition. In many cases, the proposed framework outperforms, sometimes significantly, the current state-of-the-art MIML learning methods, both in instance label prediction and bag label prediction.

Semi-supervised learning in the presence of novel class instances

In this paper, we present an approach for learning in the semi-supervised setting in the presence of novel class instances. In this setting, data consists of a labeled portion and an unlabeled portion that contains novel class instances along with unlabeled known class instances. Novel class instances are instances from concepts that do not have labeled training examples. This setting is appropriate for the case in which data is abundant and labeling the entire data is prohibitively expensive. We provide a model and an inference framework that allow for a direct control over the portion of novel class instances in the unlabeled data. Experiments on synthetic data demonstrate the usefulness of the proposed approach. Comparison to state-of-the-art approaches for learning in the presence of novel class instances using unlabeled data illustrates the advantage in using the proposed method in term of accuracy.

Simultaneous instance annotation and clustering in multi-instance multi-label learning

Multi-instance multi-label learning (MIML) is a framework that addresses label ambiguity when data contains bags, each bag contains instances, and a bag label set is provided for each bag. Instance annotation in the MIML setting is the problem of finding an instance level classifier given training data consisting of labeled bags of instances. Current approaches for instance annotation mainly focus on identifying a class label for each instance without considering inner clusters within each class. Simultaneously learning to annotate and cluster may not only yield better model fit but also help to discovery cluster structure inside each class for future investigation. This paper addresses the challenge of simultaneously annotating and clustering by proposing a graphical model that takes into account inner clusters within each class. An expectation maximization inference based on maximum likelihood is proposed for the model. Results on bird song, image annotation, and two synthetic datasets illustrate the effectiveness of the proposed framework compared to current state-of-the-art approaches.

Kernel-based instance annotation in multi-instance multi-label learning

Multi instance multi label learning is a framework in which objects are represented as bags of instances and labels are provided at the bag level. Instance annotation is the problem of assigning labels to the instances in a bag given only the bag label. Recently, OR-ed logistic regression (OR-LR) model and an EM based inference method have been proposed for instance annotation. Due to the linear nature of the logistic regression function, OR-LR performance on linearly inseparable data is limited. This paper addresses this problem by proposing a regularized kernel-based extension to the OR-LR framework. Experiments show that the kernel-based OR-LR algorithm achieves a significant improvement in classification accuracy over the linear OR-LR from 3% to 9% on audio bird song and image annotation datasets and two synthetic datasets.