Dumitru Erhan

Bio: Dumitru Erhan is an academic researcher from Google. The author has contributed to research in topics: Artificial neural network & Deep learning. The author has an hindex of 49, co-authored 68 publications receiving 87031 citations. Previous affiliations of Dumitru Erhan include Yahoo! & Microsoft.

Understanding deep architectures and the effect of unsupervised pre-training

Abstract: This thesis studies a class of algorithms called deep architectures. We argue that models that are based on a shallow composition of local features are not appropriate for the set of real-world functions and datasets that are of interest to us, namely data with many factors of variation. Modelling such functions and datasets is important if we are hoping to create an intelligent agent that can learn from complicated data. Deep architectures are hypothesized to be a step in the right direction, as they are compositions of nonlinearities and can learn compact distributed representations of data with many factors of variation. Training fully-connected artificial neural networks—the most common form of a deep architecture—was not possible before Hinton (2006) showed that one can use stacks of unsupervised Restricted Boltzmann Machines to initialize or pre-train a supervised multi-layer network. This breakthrough has been influential, as the basic idea of using unsupervised learning to improve generalization in deep networks has been reproduced in a multitude of other settings and models. In this thesis, we cast the deep learning ideas and techniques as defining a special kind of inductive bias. This bias is defined not only by the kind of functions that are eventually represented by such deep models, but also by the learning process that is commonly used for them. This work is a study of the reasons for why this class of functions generalizes well, the situations where they should work well, and the qualitative statements that one could make about such functions. This thesis is thus an attempt to understand why deep architectures work. In the first of the articles presented we study the question of how well our intuitions about the need for deep models correspond to functions that they can actually model well. In the second article we perform an in-depth study of why unsupervised pre-training helps deep learning and explore a variety of hypotheses that give us an intuition for the dynamics of learning in such architectures. Finally, in the third article, we want to better understand what a deep architecture models, qualitatively speaking. Our visualization approach enables us to understand the representations and invariances modelled and learned by deeper layers. Keywords: machine learning, artificial neural networks, deep architectures, unsupervised learning, visualization.

SurfelGAN: Synthesizing Realistic Sensor Data for Autonomous Driving

Abstract: Autonomous driving system development is critically dependent on the ability to replay complex and diverse traffic scenarios in simulation. In such scenarios, the ability to accurately simulate the vehicle sensors such as cameras, lidar or radar is essential. However, current sensor simulators leverage gaming engines such as Unreal or Unity, requiring manual creation of environments, objects and material properties. Such approaches have limited scalability and fail to produce realistic approximations of camera, lidar, and radar data without significant additional work. In this paper, we present a simple yet effective approach to generate realistic scenario sensor data, based only on a limited amount of lidar and camera data collected by an autonomous vehicle. Our approach uses texture-mapped surfels to efficiently reconstruct the scene from an initial vehicle pass or set of passes, preserving rich information about object 3D geometry and appearance, as well as the scene conditions. We then leverage a SurfelGAN network to reconstruct realistic camera images for novel positions and orientations of the self-driving vehicle and moving objects in the scene. We demonstrate our approach on the Waymo Open Dataset and show that it can synthesize realistic camera data for simulated scenarios. We also create a novel dataset that contains cases in which two self-driving vehicles observe the same scene at the same time. We use this dataset to provide additional evaluation and demonstrate the usefulness of our SurfelGAN model.

On Trade-offs of Image Prediction in Visual Model-Based Reinforcement Learning

Abstract: Model-based reinforcement learning (MBRL) methods have shown strong sample efficiency and performance across a variety of tasks, including when faced with high-dimensional visual observations These methods learn to predict the environment dynamics and expected reward from interaction and use this predictive model to plan and perform the task However, MBRL methods vary in their fundamental design choices, and it there is no strong consensus in the literature on how these design decisions affect performance In this paper, we study a number of design decisions for the predictive model in visual MBRL algorithms, focusing specifically on methods that use a predictive model for planning We find that a range of design decisions that are often considered crucial, such as the use of latent spaces, have little effect on task performance A big exception to this finding is that predicting future observations (ie, images) leads to significant task performance improvement compared to only predicting rewards We also empirically find that image prediction accuracy, somewhat surprisingly, correlates more strongly with downstream task performance than reward prediction accuracy We show how this phenomenon is related to exploration and how some of the lower-scoring models on standard benchmarks (that require exploration) will perform the same as the best-performing models when trained on the same training data Simultaneously, in the absence of exploration, models that fit the data better usually perform better on the down-stream task as well, but surprisingly, these are often not the same models that perform the best when learning and exploring from scratch These findings suggest that performance and exploration place important and potentially contradictory requirements on the model

FitVid: Overfitting in Pixel-Level Video Prediction.

Abstract: An agent that is capable of predicting what happens next can perform a variety of tasks through planning with no additional training. Furthermore, such an agent can internally represent the complex dynamics of the real-world and therefore can acquire a representation useful for a variety of visual perception tasks. This makes predicting the future frames of a video, conditioned on the observed past and potentially future actions, an interesting task which remains exceptionally challenging despite many recent advances. Existing video prediction models have shown promising results on simple narrow benchmarks but they generate low quality predictions on real-life datasets with more complicated dynamics or broader domain. There is a growing body of evidence that underfitting on the training data is one of the primary causes for the low quality predictions. In this paper, we argue that the inefficient use of parameters in the current video models is the main reason for underfitting. Therefore, we introduce a new architecture, named FitVid, which is capable of severe overfitting on the common benchmarks while having similar parameter count as the current state-of-the-art models. We analyze the consequences of overfitting, illustrating how it can produce unexpected outcomes such as generating high quality output by repeating the training data, and how it can be mitigated using existing image augmentation techniques. As a result, FitVid outperforms the current state-of-the-art models across four different video prediction benchmarks on four different metrics.

Deep Residual Learning for Image Recognition

Abstract: Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

Very Deep Convolutional Networks for Large-Scale Image Recognition

Abstract: In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.

Very Deep Convolutional Networks for Large-Scale Image Recognition

Abstract: In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.

Deep Residual Learning for Image Recognition

Abstract: Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers---8x deeper than VGG nets but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

Going deeper with convolutions

Abstract: We propose a deep convolutional neural network architecture codenamed Inception that achieves the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC14). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. By a carefully crafted design, we increased the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC14 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.