Paper Group NANR 113
SLM Lab: A Comprehensive Benchmark and Modular Software Framework for Reproducible Deep Reinforcement Learning. High-Frequency guided Curriculum Learning for Class-specific Object Boundary Detection. V1Net: A computational model of cortical horizontal connections. Hyperparameter Tuning and Implicit Regularization in Minibatch SGD. Quantum Optical E …
SLM Lab: A Comprehensive Benchmark and Modular Software Framework for Reproducible Deep Reinforcement Learning
Title | SLM Lab: A Comprehensive Benchmark and Modular Software Framework for Reproducible Deep Reinforcement Learning |
Authors | Anonymous |
Abstract | We introduce SLM Lab, a software framework for reproducible reinforcement learning (RL) research. SLM Lab implements a number of popular RL algorithms, provides synchronous and asynchronous parallel experiment execution, hyperparameter search, and result analysis. RL algorithms in SLM Lab are implemented in a modular way such that differences in algorithm performance can be confidently ascribed to differences between algorithms, not between implementations. In this work we present the design choices behind SLM Lab and use it to produce a comprehensive single-codebase RL algorithm benchmark. In addition, as a consequence of SLM Lab’s modular design, we introduce and evaluate a discrete-action variant of the Soft Actor-Critic algorithm (Haarnoja et al., 2018) and a hybrid synchronous/asynchronous training method for RL agents. |
Tasks | |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=S1gLBgBtDH |
https://openreview.net/pdf?id=S1gLBgBtDH | |
PWC | https://paperswithcode.com/paper/slm-lab-a-comprehensive-benchmark-and-modular |
Repo | |
Framework | |
High-Frequency guided Curriculum Learning for Class-specific Object Boundary Detection
Title | High-Frequency guided Curriculum Learning for Class-specific Object Boundary Detection |
Authors | Anonymous |
Abstract | This work addresses class-specific object boundary extraction, i.e., retrieving boundary pixels that belong to a class of objects in the given image. Although recent ConvNet-based approaches demonstrate impressive results, we notice that they produce several false-alarms and misdetections when used in real-world applications. We hypothesize that although boundary detection is simple at some pixels that are rooted in identifiable high-frequency locations, other pixels pose a higher level of difficulties, for instance, region pixels with an appearance similar to the boundaries; or boundary pixels with insignificant edge strengths. Therefore, the training process needs to account for different levels of learning complexity in different regions to overcome false alarms. In this work, we devise a curriculum-learning-based training process for object boundary detection. This multi-stage training process first trains the network at simpler pixels (with sufficient edge strengths) and then at harder pixels in the later stages of the curriculum. We also propose a novel system for object boundary detection that relies on a fully convolutional neural network (FCN) and wavelet decomposition of image frequencies. This system uses high-frequency bands from the wavelet pyramid and augments them to conv features from different layers of FCN. Our ablation studies with contourMNIST dataset, a simulated digit contours from MNIST, demonstrate that this explicit high-frequency augmentation helps the model to converge faster. Our model trained by the proposed curriculum scheme outperforms a state-of-the-art object boundary detection method by a significant margin on a challenging aerial image dataset. |
Tasks | Boundary Detection |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=HkeMYJHYvS |
https://openreview.net/pdf?id=HkeMYJHYvS | |
PWC | https://paperswithcode.com/paper/high-frequency-guided-curriculum-learning-for |
Repo | |
Framework | |
V1Net: A computational model of cortical horizontal connections
Title | V1Net: A computational model of cortical horizontal connections |
Authors | Anonymous |
Abstract | The primate visual system builds robust, multi-purpose representations of the external world in order to support several diverse downstream cortical processes. Such representations are required to be invariant to the sensory inconsistencies caused by dynamically varying lighting, local texture distortion, etc. A key architectural feature combating such environmental irregularities is ‘long-range horizontal connections’ that aid the perception of the global form of objects. In this work, we explore the introduction of such horizontal connections into standard deep convolutional networks; we present V1Net – a novel convolutional-recurrent unit that models linear and nonlinear horizontal inhibitory and excitatory connections inspired by primate visual cortical connectivity. We introduce the Texturized Challenge – a new benchmark to evaluate object recognition performance under perceptual noise – which we use to evaluate V1Net against an array of carefully selected control models with/without recurrent processing. Additionally, we present results from an ablation study of V1Net demonstrating the utility of diverse neurally inspired horizontal connections for state-of-the-art AI systems on the task of object boundary detection from natural images. We also present the emergence of several biologically plausible horizontal connectivity patterns, namely center-on surround-off, association fields and border-ownership connectivity patterns in a V1Net model trained to perform boundary detection on natural images from the Berkeley Segmentation Dataset 500 (BSDS500). Our findings suggest an increased representational similarity between V1Net and biological visual systems, and highlight the importance of neurally inspired recurrent contextual processing principles for learning visual representations that are robust to perceptual noise and furthering the state-of-the-art in computer vision. |
Tasks | Boundary Detection, Object Recognition |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=Hyg4kkHKwH |
https://openreview.net/pdf?id=Hyg4kkHKwH | |
PWC | https://paperswithcode.com/paper/v1net-a-computational-model-of-cortical |
Repo | |
Framework | |
Hyperparameter Tuning and Implicit Regularization in Minibatch SGD
Title | Hyperparameter Tuning and Implicit Regularization in Minibatch SGD |
Authors | Anonymous |
Abstract | This paper makes two contributions towards understanding how the hyperparameters of stochastic gradient descent affect the final training loss and test accuracy of neural networks. First, we argue that stochastic gradient descent exhibits two regimes with different behaviours; a noise dominated regime which typically arises for small or moderate batch sizes, and a curvature dominated regime which typically arises when the batch size is large. In the noise dominated regime, the optimal learning rate increases as the batch size rises, and the training loss and test accuracy are independent of batch size under a constant epoch budget. In the curvature dominated regime, the optimal learning rate is independent of batch size, and the training loss and test accuracy degrade as the batch size rises. We support these claims with experiments on a range of architectures including ResNets, LSTMs and autoencoders. We always perform a grid search over learning rates at all batch sizes. Second, we demonstrate that small or moderately large batch sizes continue to outperform very large batches on the test set, even when both models are trained for the same number of steps and reach similar training losses. Furthermore, when training Wide-ResNets on CIFAR-10 with a constant batch size of 64, the optimal learning rate to maximize the test accuracy only decays by a factor of 2 when the epoch budget is increased by a factor of 128, while the optimal learning rate to minimize the training loss decays by a factor of 16. These results confirm that the noise in stochastic gradients can introduce beneficial implicit regularization. |
Tasks | |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=ryGWhJBtDB |
https://openreview.net/pdf?id=ryGWhJBtDB | |
PWC | https://paperswithcode.com/paper/hyperparameter-tuning-and-implicit |
Repo | |
Framework | |
Quantum Optical Experiments Modeled by Long Short-Term Memory
Title | Quantum Optical Experiments Modeled by Long Short-Term Memory |
Authors | Anonymous |
Abstract | We demonstrate how machine learning is able to model experiments in quantum physics. Quantum entanglement is a cornerstone for upcoming quantum technologies such as quantum computation and quantum cryptography. Of particular interest are complex quantum states with more than two particles and a large number of entangled quantum levels. Given such a multiparticle high-dimensional quantum state, it is usually impossible to reconstruct an experimental setup that produces it. To search for interesting experiments, one thus has to randomly create millions of setups on a computer and calculate the respective output states. In this work, we show that machine learning models can provide significant improvement over random search. We demonstrate that a long short-term memory (LSTM) neural network can successfully learn to model quantum experiments by correctly predicting output state characteristics for given setups without the necessity of computing the states themselves. This approach not only allows for faster search but is also an essential step towards automated design of multiparticle high-dimensional quantum experiments using generative machine learning models. |
Tasks | |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=ryxtWgSKPB |
https://openreview.net/pdf?id=ryxtWgSKPB | |
PWC | https://paperswithcode.com/paper/quantum-optical-experiments-modeled-by-long |
Repo | |
Framework | |
Variational Hashing-based Collaborative Filtering with Self-Masking
Title | Variational Hashing-based Collaborative Filtering with Self-Masking |
Authors | Anonymous |
Abstract | Hashing-based collaborative filtering learns binary vector representations (hash codes) of users and items, such that recommendations can be computed very efficiently using the Hamming distance, which is simply the sum of differing bits between two hash codes. A problem with hashing-based collaborative filtering using the Hamming distance, is that each bit is equally weighted in the distance computation, but in practice some bits might encode more important properties than other bits, where the importance depends on the user. To this end, we propose an end-to-end trainable variational hashing-based collaborative filtering approach that uses the novel concept of self-masking: the user hash code acts as a mask on the items (using the Boolean AND operation), such that it learns to encode which bits are important to the user, rather than the user’s preference towards the underlying item property that the bits represent. This allows a binary user-level importance weighting of each item without the need to store additional weights for each user. We experimentally evaluate our approach against state-of-the-art baselines on 4 datasets, and obtain significant gains of up to 12% in NDCG. We also make available an efficient implementation of self-masking, which experimentally yields <4% runtime overhead compared to the standard Hamming distance. |
Tasks | |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=rylDzTEKwr |
https://openreview.net/pdf?id=rylDzTEKwr | |
PWC | https://paperswithcode.com/paper/variational-hashing-based-collaborative |
Repo | |
Framework | |
Batch Normalization is a Cause of Adversarial Vulnerability
Title | Batch Normalization is a Cause of Adversarial Vulnerability |
Authors | Anonymous |
Abstract | Batch normalization (BN) is often used in an attempt to stabilize and accelerate training in deep neural networks. In many cases it indeed decreases the number of parameter updates required to achieve low training error. However, it also reduces robustness to small adversarial input perturbations and common corruptions by double-digit percentages, as we show on five standard datasets. Furthermore, we find that substituting weight decay for BN is sufficient to nullify a relationship between adversarial vulnerability and the input dimension. A recent mean-field analysis found that BN induces gradient explosion when used on multiple layers, but this cannot fully explain the vulnerability we observe, given that it occurs already for a single BN layer. We argue that the actual cause is the tilting of the decision boundary with respect to the nearest-centroid classifier along input dimensions of low variance. As a result, the constant introduced for numerical stability in the BN step acts as an important hyperparameter that can be tuned to recover some robustness at the cost of standard test accuracy. We explain this mechanism explicitly on a linear toy model and show in experiments that it still holds for nonlinear real-world models. |
Tasks | |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=H1x-3xSKDr |
https://openreview.net/pdf?id=H1x-3xSKDr | |
PWC | https://paperswithcode.com/paper/batch-normalization-is-a-cause-of-adversarial-1 |
Repo | |
Framework | |
Enhancing Language Emergence through Empathy
Title | Enhancing Language Emergence through Empathy |
Authors | Marie Ossenkopf |
Abstract | The emergence of language in multi-agent settings is a promising research direction to ground natural language in simulated agents. If AI would be able to understand the meaning of language through its using it, it could also transfer it to other situations flexibly. That is seen as an important step towards achieving general AI. The scope of emergent communication is so far, however, still limited. It is necessary to enhance the learning possibilities for skills associated with communication to increase the emergable complexity. We took an example from human language acquisition and the importance of the empathic connection in this process. We propose an approach to introduce the notion of empathy to multi-agent deep reinforcement learning. We extend existing approaches on referential games with an auxiliary task for the speaker to predict the listener’s mind change improving the learning time. Our experiments show the high potential of this architectural element by doubling the learning speed of the test setup. |
Tasks | Language Acquisition |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=Hke1gySFvB |
https://openreview.net/pdf?id=Hke1gySFvB | |
PWC | https://paperswithcode.com/paper/enhancing-language-emergence-through-empathy |
Repo | |
Framework | |
The Role of Embedding Complexity in Domain-invariant Representations
Title | The Role of Embedding Complexity in Domain-invariant Representations |
Authors | Anonymous |
Abstract | Unsupervised domain adaptation aims to generalize the hypothesis trained in a source domain to an unlabeled target domain. One popular approach to this problem is to learn domain-invariant embeddings for both domains. In this work, we study, theoretically and empirically, the effect of the embedding complexity on generalization to the target domain. In particular, this complexity affects an upper bound on the target risk; this is reflected in experiments, too. Next, we specify our theoretical framework to multilayer neural networks. As a result, we develop a strategy that mitigates sensitivity to the embedding complexity, and empirically achieves performance on par with or better than the best layer-dependent complexity tradeoff. |
Tasks | Domain Adaptation, Unsupervised Domain Adaptation |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=SkgOzlrKvH |
https://openreview.net/pdf?id=SkgOzlrKvH | |
PWC | https://paperswithcode.com/paper/the-role-of-embedding-complexity-in-domain |
Repo | |
Framework | |
Training Data Distribution Search with Ensemble Active Learning
Title | Training Data Distribution Search with Ensemble Active Learning |
Authors | Anonymous |
Abstract | Deep Neural Networks (DNNs) often rely on very large datasets for training. Given the large size of such datasets, it is conceivable that they contain certain samples that either do not contribute or negatively impact the DNN’s optimization. Modifying the training distribution in a way that excludes such samples could provide an effective solution to both improve performance and reduce training time. In this paper, we propose to scale up ensemble Active Learning methods to perform acquisition at a large scale (10k to 500k samples at a time). We do this with ensembles of hundreds of models, obtained at a minimal computational cost by reusing intermediate training checkpoints. This allows us to automatically and efficiently perform a training data distribution search for large labeled datasets. We observe that our approach obtains favorable subsets of training data, which can be used to train more accurate DNNs than training with the entire dataset. We perform an extensive experimental study of this phenomenon on three image classification benchmarks (CIFAR-10, CIFAR-100 and ImageNet), analyzing the impact of initialization schemes, acquisition functions and ensemble configurations. We demonstrate that data subsets identified with a lightweight ResNet-18 ensemble remain effective when used to train deep models like ResNet-101 and DenseNet-121. Our results provide strong empirical evidence that optimizing the training data distribution can provide significant benefits on large scale vision tasks. |
Tasks | Active Learning, Image Classification |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=rkxEKp4Fwr |
https://openreview.net/pdf?id=rkxEKp4Fwr | |
PWC | https://paperswithcode.com/paper/training-data-distribution-search-with |
Repo | |
Framework | |
TWIN GRAPH CONVOLUTIONAL NETWORKS: GCN WITH DUAL GRAPH SUPPORT FOR SEMI-SUPERVISED LEARNING
Title | TWIN GRAPH CONVOLUTIONAL NETWORKS: GCN WITH DUAL GRAPH SUPPORT FOR SEMI-SUPERVISED LEARNING |
Authors | Anonymous |
Abstract | Graph Neural Networks as a combination of Graph Signal Processing and Deep Convolutional Networks shows great power in pattern recognition in non-Euclidean domains. In this paper, we propose a new method to deploy two pipelines based on the duality of a graph to improve accuracy. By exploring the primal graph and its dual graph where nodes and edges can be treated as one another, we have exploited the benefits of both vertex features and edge features. As a result, we have arrived at a framework that has great potential in both semisupervised and unsupervised learning. |
Tasks | |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=SkxV7kHKvr |
https://openreview.net/pdf?id=SkxV7kHKvr | |
PWC | https://paperswithcode.com/paper/twin-graph-convolutional-networks-gcn-with |
Repo | |
Framework | |
Learning to Infer User Interface Attributes from Images
Title | Learning to Infer User Interface Attributes from Images |
Authors | Anonymous |
Abstract | We present a new approach that helps developers automate the process of user interface implementation. Concretely, given an input image created by a designer (e.g, using a vector graphics editor), we learn to infer its implementation which when rendered (e.g., on the Android platform), looks visually the same as the input image. To achieve this, we take a black box rendering engine and a set of attributes it supports (e.g., colors, border radius, shadow or text properties), use it to generate a suitable synthetic training dataset, and then train specialized neural models to predict each of the attribute values. To improve pixel-level accuracy, we also use imitation learning to train a neural policy that refines the predicted attribute values by learning to compute the similarity of the original and rendered images in their attribute space, rather than based on the difference of pixel values. |
Tasks | Imitation Learning |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=rylNJlStwB |
https://openreview.net/pdf?id=rylNJlStwB | |
PWC | https://paperswithcode.com/paper/learning-to-infer-user-interface-attributes |
Repo | |
Framework | |
Stochastic Neural Physics Predictor
Title | Stochastic Neural Physics Predictor |
Authors | Anonymous |
Abstract | Recently, neural-network based forward dynamics models have been proposed that attempt to learn the dynamics of physical systems in a deterministic way. While near-term motion can be predicted accurately, long-term predictions suffer from accumulating input and prediction errors which can lead to plausible but different trajectories that diverge from the ground truth. A system that predicts distributions of the future physical states for long time horizons based on its uncertainty is thus a promising solution. In this work, we introduce a novel robust Monte Carlo sampling based graph-convolutional dropout method that allows us to sample multiple plausible trajectories for an initial state given a neural-network based forward dynamics predictor. By introducing a new shape preservation loss and training our dynamics model recurrently, we stabilize long-term predictions. We show that our model’s long-term forward dynamics prediction errors on complicated physical interactions of rigid and deformable objects of various shapes are significantly lower than existing strong baselines. Lastly, we demonstrate how generating multiple trajectories with our Monte Carlo dropout method can be used to train model-free reinforcement learning agents faster and to better solutions on simple manipulation tasks. |
Tasks | |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=HkgXteBYPB |
https://openreview.net/pdf?id=HkgXteBYPB | |
PWC | https://paperswithcode.com/paper/stochastic-neural-physics-predictor |
Repo | |
Framework | |
Point Process Flows
Title | Point Process Flows |
Authors | Anonymous |
Abstract | Event sequences can be modeled by temporal point processes (TPPs) to capture their asynchronous and probabilistic nature. We propose an intensity-free framework that directly models the point process as a non-parametric distribution by utilizing normalizing flows. This approach is capable of capturing highly complex temporal distributions and does not rely on restrictive parametric forms. Comparisons with state-of-the-art baseline models on both synthetic and challenging real-life datasets show that the proposed framework is effective at modeling the stochasticity of discrete event sequences. |
Tasks | Point Processes |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=rklJ2CEYPH |
https://openreview.net/pdf?id=rklJ2CEYPH | |
PWC | https://paperswithcode.com/paper/point-process-flows-1 |
Repo | |
Framework | |
Spectral Nonlocal Block for Neural Network
Title | Spectral Nonlocal Block for Neural Network |
Authors | Anonymous |
Abstract | The nonlocal network is designed for capturing long-range spatial-temporal dependencies in several computer vision tasks. Although having shown excellent performances, it needs an elaborate preparation for both the number and position of the building blocks. In this paper, we propose a new formulation of the nonlocal block and interpret it from the general graph signal processing perspective, where we view it as a fully-connected graph filter approximated by Chebyshev polynomials. The proposed nonlocal block is more efficient and robust, which is a generalized form of existing nonlocal blocks (e.g. nonlocal block, nonlocal stage). Moreover, we give the stable hypothesis and show that the steady-state of the deeper nonlocal structure should meet with it. Based on the stable hypothesis, a full-order approximation of the nonlocal block is derived for consecutive connections. Experimental results illustrate the clear-cut improvement and practical applicability of the generalized nonlocal block on both image and video classification tasks. |
Tasks | Video Classification |
Published | 2020-01-01 |
URL | https://openreview.net/forum?id=rkgb9kSKwS |
https://openreview.net/pdf?id=rkgb9kSKwS | |
PWC | https://paperswithcode.com/paper/spectral-nonlocal-block-for-neural-network |
Repo | |
Framework | |