Honglak Lee

Convolutional deep belief networks for scalable unsupervised learning of hierarchical representations

There has been much interest in unsupervised learning of hierarchical generative models such as deep belief networks. Scaling such models to full-sized, high-dimensional images remains a difficult problem. To address this problem, we present the convolutional deep belief network, a hierarchical generative model which scales to realistic image sizes. This model is translation-invariant and supports efficient bottom-up and top-down probabilistic inference. Key to our approach is probabilistic max-pooling, a novel technique which shrinks the representations of higher layers in a probabilistically sound way. Our experiments show that the algorithm learns useful high-level visual features, such as object parts, from unlabeled images of objects and natural scenes. We demonstrate excellent performance on several visual recognition tasks and show that our model can perform hierarchical (bottom-up and top-down) inference over full-sized images.

Deep learning for detecting robotic grasps

We consider the problem of detecting robotic grasps in an RGB-D view of a scene containing objects. In this work, we apply a deep learning approach to solve this problem, which avoids time-consuming hand-design of features. This presents two main challenges. First, we need to evaluate a huge number of candidate grasps. In order to make detection fast and robust, we present a two-step cascaded system with two deep networks, where the top detections from the first are re-evaluated by the second. The first network has fewer features, is faster to run, and can effectively prune out unlikely candidate grasps. The second, with more features, is slower but has to run only on the top few detections. Second, we need to handle multimodal inputs effectively, for which we present a method that applies structured regularization on the weights based on multimodal group regularization. We show that our method improves performance on an RGBD robotic grasping dataset, and can be used to successfully execute grasps on two different robotic platforms.

Learning hierarchical representations for face verification with convolutional deep belief networks

Most modern face recognition systems rely on a feature representation given by a hand-crafted image descriptor, such as Local Binary Patterns (LBP), and achieve improved performance by combining several such representations. In this paper, we propose deep learning as a natural source for obtaining additional, complementary representations. To learn features in high-resolution images, we make use of convolutional deep belief networks. Moreover, to take advantage of global structure in an object class, we develop local convolutional restricted Boltzmann machines, a novel convolutional learning model that exploits the global structure by not assuming stationarity of features across the image, while maintaining scalability and robustness to small misalignments. We also present a novel application of deep learning to descriptors other than pixel intensity values, such as LBP. In addition, we compare performance of networks trained using unsupervised learning against networks with random filters, and empirically show that learning weights not only is necessary for obtaining good multilayer representations, but also provides robustness to the choice of the network architecture parameters. Finally, we show that a recognition system using only representations obtained from deep learning can achieve comparable accuracy with a system using a combination of hand-crafted image descriptors. Moreover, by combining these representations, we achieve state-of-the-art results on a real-world face verification database.

Unsupervised learning of hierarchical representations with convolutional deep belief networks

There has been much interest in unsupervised learning of hierarchical generative models such as deep belief networks (DBNs); however, scaling such models to full-sized, high-dimensional images remains a difficult problem. To address this problem, we present the convolutional deep belief network, a hierarchical generative model that scales to realistic image sizes. This model is translation-invariant and supports efficient bottom-up and top-down probabilistic inference. Key to our approach is probabilistic max-pooling, a novel technique that shrinks the representations of higher layers in a probabilistically sound way. Our experiments show that the algorithm learns useful high-level visual features, such as object parts, from unlabeled images of objects and natural scenes. We demonstrate excellent performance on several visual recognition tasks and show that our model can perform hierarchical (bottom-up and top-down) inference over full-sized images.

Attribute2Image: Conditional Image Generation from Visual Attributes

This paper investigates a novel problem of generating images from visual attributes. We model the image as a composite of foreground and background and develop a layered generative model with disentangled latent variables that can be learned end-to-end using a variational auto-encoder. We experiment with natural images of faces and birds and demonstrate that the proposed models are capable of generating realistic and diverse samples with disentangled latent representations. We use a general energy minimization algorithm for posterior inference of latent variables given novel images. Therefore, the learned generative models show excellent quantitative and visual results in the tasks of attribute-conditioned image reconstruction and completion.

High-throughput identification of transcription start sites, conserved promoter motifs and predicted regulons

Using 62 probe-level datasets obtained with a custom-designed Caulobacter crescentus microarray chip, we identify transcriptional start sites of 769 genes, 53 of which are transcribed from multiple start sites. Transcriptional start sites are identified by analyzing probe signal cross-correlation matrices created from probe pairs tiled every 5 bp upstream of the genes. Signals from probes binding the same message are correlated. The contribution of each promoter for genes transcribed from multiple promoters is identified. Knowing the transcription start site enables targeted searching for regulatory-protein binding motifs in the promoter regions of genes with similar expression patterns. We identified 27 motifs, 17 of which share no similarity to the characterized motifs of other C. crescentus transcriptional regulators. Using these motifs, we predict coregulated genes. We verified novel promoter motifs that regulate stress-response genes, including those responding to uranium challenge, a stress-response sigma factor and a stress-response noncoding RNA.

Deep learning for robust feature generation in audiovisual emotion recognition

Automatic emotion recognition systems predict high-level affective content from low-level human-centered signal cues. These systems have seen great improvements in classification accuracy, due in part to advances in feature selection methods. However, many of these feature selection methods capture only linear relationships between features or alternatively require the use of labeled data. In this paper we focus on deep learning techniques, which can overcome these limitations by explicitly capturing complex non-linear feature interactions in multimodal data. We propose and evaluate a suite of Deep Belief Network models, and demonstrate that these models show improvement in emotion classification performance over baselines that do not employ deep learning. This suggests that the learned high-order non-linear relationships are effective for emotion recognition.

Evaluation of output embeddings for fine-grained image classification

Image classification has advanced significantly in recent years with the availability of large-scale image sets. However, fine-grained classification remains a major challenge due to the annotation cost of large numbers of fine-grained categories. This project shows that compelling classification performance can be achieved on such categories even without labeled training data. Given image and class embeddings, we learn a compatibility function such that matching embeddings are assigned a higher score than mismatching ones; zero-shot classification of an image proceeds by finding the label yielding the highest joint compatibility score. We use state-of-the-art image features and focus on different supervised attributes and unsupervised output embeddings either derived from hierarchies or learned from unlabeled text corpora. We establish a substantially improved state-of-the-art on the Animals with Attributes and Caltech-UCSD Birds datasets. Most encouragingly, we demonstrate that purely unsupervised output embeddings (learned from Wikipedia and improved with finegrained text) achieve compelling results, even outperforming the previous supervised state-of-the-art. By combining different output embeddings, we further improve results.Despite significant recent advances in image classification, fine-grained classification remains a challenge. In the present paper, we address the zero-shot and few-shot learning scenarios as obtaining labeled data is especially difficult for fine-grained classification tasks. First, we embed state-of-the-art image descriptors in a label embedding space using side information such as attributes. We argue that learning a joint embedding space, that maximizes the compatibility between the input and output embeddings, is highly effective for zero/few-shot learning. We show empirically that such embeddings significantly outperforms the current state-of-the-art methods on two challenging datasets (Caltech-UCSD Birds and Animals with Attributes). Second, to reduce the amount of costly manual attribute annotations, we use alternate output embeddings based on the word-vector representations, obtained from large text-corpora without any supervision. We report that such unsupervised embeddings achieve encouraging results, and lead to further improvements when combined with the supervised ones.

Efficient learning of sparse, distributed, convolutional feature representations for object recognition

Informative image representations are important in achieving state-of-the-art performance in object recognition tasks. Among feature learning algorithms that are used to develop image representations, restricted Boltzmann machines (RBMs) have good expressive power and build effective representations. However, the difficulty of training RBMs has been a barrier to their wide use. To address this difficulty, we show the connections between mixture models and RBMs and present an efficient training method for RBMs that utilize these connections. To the best of our knowledge, this is the first work showing that RBMs can be trained with almost no hyperparameter tuning to provide classification performance similar to or significantly better than mixture models (e.g., Gaussian mixture models). Along with this efficient training, we evaluate the importance of convolutional training that can capture a larger spatial context with less redundancy, as compared to non-convolutional training. Overall, our method achieves state-of-the-art performance on both Caltech 101 / 256 datasets using a single type of feature.

Improving object detection with deep convolutional networks via Bayesian optimization and structured prediction

Object detection systems based on the deep convolutional neural network (CNN) have recently made ground-breaking advances on several object detection benchmarks. While the features learned by these high-capacity neural networks are discriminative for categorization, inaccurate localization is still a major source of error for detection. Building upon high-capacity CNN architectures, we address the localization problem by 1) using a search algorithm based on Bayesian optimization that sequentially proposes candidate regions for an object bounding box, and 2) training the CNN with a structured loss that explicitly penalizes the localization inaccuracy. In experiments, we demonstrate that each of the proposed methods improves the detection performance over the baseline method on PASCAL VOC 2007 and 2012 datasets. Furthermore, two methods are complementary and significantly outperform the previous state-of-the-art when combined.