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.

/pdf/going-deeper-with-convolutions-1yobw2o2ds.pdf

Going deeper with convolutions

Deep neural nets with a large number of parameters are very powerful machine learning systems. However, overfitting is a serious problem in such networks. Large networks are also slow to use, making it difficult to deal with overfitting by combining the predictions of many different large neural nets at test time. Dropout is a technique for addressing this problem. The key idea is to randomly drop units (along with their connections) from the neural network during training. This prevents units from co-adapting too much. During training, dropout samples from an exponential number of different "thinned" networks. At test time, it is easy to approximate the effect of averaging the predictions of all these thinned networks by simply using a single unthinned network that has smaller weights. This significantly reduces overfitting and gives major improvements over other regularization methods. We show that dropout improves the performance of neural networks on supervised learning tasks in vision, speech recognition, document classification and computational biology, obtaining state-of-the-art results on many benchmark data sets.

/pdf/dropout-a-simple-way-to-prevent-neural-networks-from-4gtr92okhv.pdf

Dropout: a simple way to prevent neural networks from overfitting

We present a new technique called “t-SNE” that visualizes high-dimensional data by giving each datapoint a location in a two or three-dimensional map. The technique is a variation of Stochastic Neighbor Embedding (Hinton and Roweis, 2002) that is much easier to optimize, and produces significantly better visualizations by reducing the tendency to crowd points together in the center of the map. t-SNE is better than existing techniques at creating a single map that reveals structure at many different scales. This is particularly important for high-dimensional data that lie on several different, but related, low-dimensional manifolds, such as images of objects from multiple classes seen from multiple viewpoints. For visualizing the structure of very large datasets, we show how t-SNE can use random walks on neighborhood graphs to allow the implicit structure of all of the data to influence the way in which a subset of the data is displayed. We illustrate the performance of t-SNE on a wide variety of datasets and compare it with many other non-parametric visualization techniques, including Sammon mapping, Isomap, and Locally Linear Embedding. The visualizations produced by t-SNE are significantly better than those produced by the other techniques on almost all of the datasets.

/pdf/visualizing-data-using-t-sne-2w1n304iq0.pdf

Visualizing Data using t-SNE

Recent work has shown that convolutional networks can be substantially deeper, more accurate, and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper, we embrace this observation and introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections&#x2014;one between each layer and its subsequent layer&#x2014;our network has L(L+1)/2 direct connections. For each layer, the feature-maps of all preceding layers are used as inputs, and its own feature-maps are used as inputs into all subsequent layers. DenseNets have several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters. We evaluate our proposed architecture on four highly competitive object recognition benchmark tasks (CIFAR-10, CIFAR-100, SVHN, and ImageNet). DenseNets obtain significant improvements over the state-of-the-art on most of them, whilst requiring less memory and computation to achieve high performance. Code and pre-trained models are available at https://github.com/liuzhuang13/DenseNet.

/pdf/densely-connected-convolutional-networks-i745msea9d.pdf

Densely Connected Convolutional Networks

Convolutional networks are at the core of most state of-the-art computer vision solutions for a wide variety of tasks. Since 2014 very deep convolutional networks started to become mainstream, yielding substantial gains in various benchmarks. Although increased model size and computational cost tend to translate to immediate quality gains for most tasks (as long as enough labeled data is provided for training), computational efficiency and low parameter count are still enabling factors for various use cases such as mobile vision and big-data scenarios. Here we are exploring ways to scale up networks in ways that aim at utilizing the added computation as efficiently as possible by suitably factorized convolutions and aggressive regularization. We benchmark our methods on the ILSVRC 2012 classification challenge validation set demonstrate substantial gains over the state of the art: 21:2% top-1 and 5:6% top-5 error for single frame evaluation using a network with a computational cost of 5 billion multiply-adds per inference and with using less than 25 million parameters. With an ensemble of 4 models and multi-crop evaluation, we report 3:5% top-5 error and 17:3% top-1 error on the validation set and 3:6% top-5 error on the official test set.

Rethinking the Inception Architecture for Computer Vision

The generalization properties of most existing machine learning techniques are predicated on the assumptions that 1) a sufficiently large quantity of training data is available; 2) the training and testing data come from some common distribution. Although these assumptions are often met in practice, there are also many scenarios in which training data from the relevant distribution is insufficient. We focus on making use of additional data, which is readily available or can be obtained easily but comes from a different distribution than the testing data, to aid learning. 
We present five learning scenarios, depending on how the distribution we used to sample the additional training data differs from the testing distribution: 1) learning with weak supervision; 2) domain adaptation; 3) learning from multiple domains; 4) learning from corrupted data; 5) learning with partial supervision. 
We introduce two strategies and manifest them in five ways to cope with the difference between the training and testing distribution. The first strategy, which gives rise to Pseudo Multi-view Co-training (PMC) and Co-training for Domain Adaptation (CODA), is inspired by the co-training algorithm for multi-view data. PMC generalizes co-training to the more common single view data and allows us to learn from weakly labeled data retrieved free from the web. CODA integrates PMC with an another feature selection component to address the feature incompatibility between domains for domain adaptation. PMC and CODA are evaluated on a variety of real datasets, and both yield record performance. 
The second strategy marginalized dropout leads to marginalized Stacked Denoising Autoencoders (mSDA), Marginalized Corrupted Features (MCF) and FastTag (FastTag). mSDA diminishes the difference between distributions associated with different domains by learning a new representation through marginalized corruption and reconstruciton. MCF learns from a known distribution which is created by corrupting a small set of training data, and improves robustness of learned classifiers by training on "infinitely'' many data sampled from the distribution. FastTag applies marginalized dropout to the output of partially labeled data to recover missing labels for multi-label tasks. These three algorithms not only achieve the state-of-art performance in various tasks, but also deliver orders of magnitude speed up at training and testing comparing to competing algorithms.

Learning with single view co-training and marginalized dropout

Recently, a Euclidean heuristic (EH) has been proposed for A* search. EH exploits manifold learning methods to construct an embedding of the state space graph, and derives an admissible heuristic distance between two states from the Euclidean distance between their respective embedded points. EH has shown good performance and memory efficiency in comparison to other existing heuristics such as differential heuristics. However, its potential has not been fully explored. In this paper, we propose a number of techniques that can significantly improve the quality of EH. We propose a goal-oriented manifold learning scheme that optimizes the Euclidean distance to goals in the embedding while maintaining admissibility and consistency. We also propose a state heuristic enhancement technique to reduce the gap between heuristic and true distances. The enhanced heuristic is admissible but no longer consistent. We then employ a modified search algorithm, known as B′ algorithm, that achieves optimality with inconsistent heuristics using consistency check and propagation. We demonstrate the effectiveness of the above techniques and report un-matched reduction in search costs across several non-trivial benchmark search problems.

Goal-oriented Euclidean heuristics with manifold learning

User feedback is vital to the quality of the collaborative spam fi lters frequently used in open membership email systems such as Yahoo Mail or Gmail Users occasionally designate emails as spam or non-spam (often termed as ham), and these labels are subsequently used to train the spam fi lter Although the majority of users provide very little data, as a collective the amount of training data is very large (many millions of emails per day) Unfortunately, there is substantial deviation in users’ notions of what constitutes spam and ham Additionally, the open membership policy of these systems makes it vulnerable to users with malicious intent – spammers who wish to see their emails accepted by any spam fi ltration system can create accounts and use these to give malicious feedback to ‘train’ the spam fi lter in giving their emails a free pass When combined, these realities make it extremely diffi cult to assemble a single, global spam classifi er

/pdf/collaborative-spam-filtering-with-the-hashing-trick-3zmgc6d5mn.pdf

Collaborative spam filtering with the hashing trick

In this technical report, we introduce FastFusionNet, an efficient variant of FusionNet [12]. FusionNet is a high performing reading comprehension architecture, which was designed primarily for maximum retrieval accuracy with less regard towards computational requirements. For FastFusionNets we remove the expensive CoVe layers [21] and substitute the BiLSTMs with far more efficient SRU layers [19]. The resulting architecture obtains state-of-the-art results on DAWNBench [5] while achieving the lowest training and inference time on SQuAD [25] to-date. The code is available at this https URL.

FastFusionNet: New State-of-the-Art for DAWNBench SQuAD.

Recent work has shown that convolutional networks can be substantially deeper, more accurate, and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper, we embrace this observation and introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion.Whereas traditional convolutional networks with L layers have L connections - one between each layer and its subsequent layer - our network has L(L+1)/2 direct connections. For each layer, the feature-maps of all preceding layers are used as inputs, and its own feature-maps are used as inputs into all subsequent layers. DenseNets have several compelling advantages: they alleviate the vanishing-gradient problem, encourage feature reuse and substantially improve parameter efficiency. We evaluate our proposed architecture on four highly competitive object recognition benchmark tasks (CIFAR-10, CIFAR-100, SVHN, and ImageNet). DenseNets obtain significant improvements over the state-of-the-art on most of them, whilst requiring less parameters and computation to achieve high performance.

Kilian Q. Weinberger

Papers

Learning with single view co-training and marginalized dropout

Goal-oriented Euclidean heuristics with manifold learning

Collaborative spam filtering with the hashing trick

FastFusionNet: New State-of-the-Art for DAWNBench SQuAD.

Convolutional Networks with Dense Connectivity