We show that training of generative adversarial network (GAN) may not have good generalization properties; e.g., training may appear successful but the trained distribution may be far from target distribution in standard metrics. However, generalization does occur for a weaker metric called neural net distance. It is also shown that an approximate pure equilibrium exists in the discriminator/generator game for a special class of generators with natural training objectives when generator capacity and training set sizes are moderate. This existence of equilibrium inspires MIX+GAN protocol, which can be combined with any existing GAN training, and empirically shown to improve some of them.
https://arxiv.org/abs/1703.00573
Generative Adversarial Networks (GANs) have recently achieved significant improvement on paired/unpaired image-to-image translation, such as photo$\rightarrow$ sketch and artist painting style transfer. However, existing models can only be capable of transferring the low-level information (e.g. color or texture changes), but fail to edit high-level semantic meanings (e.g., geometric structure or content) of objects. On the other hand, while some researches can synthesize compelling real-world images given a class label or caption, they cannot condition on arbitrary shapes or structures, which largely limits their application scenarios and interpretive capability of model results. In this work, we focus on a more challenging semantic manipulation task, which aims to modify the semantic meaning of an object while preserving its own characteristics (e.g. viewpoints and shapes), such as cow$\rightarrow$sheep, motor$\rightarrow$ bicycle, cat$\rightarrow$dog. To tackle such large semantic changes, we introduce a contrasting GAN (contrast-GAN) with a novel adversarial contrasting objective. Instead of directly making the synthesized samples close to target data as previous GANs did, our adversarial contrasting objective optimizes over the distance comparisons between samples, that is, enforcing the manipulated data be semantically closer to the real data with target category than the input data. Equipped with the new contrasting objective, a novel mask-conditional contrast-GAN architecture is proposed to enable disentangle image background with object semantic changes. Experiments on several semantic manipulation tasks on ImageNet and MSCOCO dataset show considerable performance gain by our contrast-GAN over other conditional GANs. Quantitative results further demonstrate the superiority of our model on generating manipulated results with high visual fidelity and reasonable object semantics.
https://arxiv.org/abs/1708.00315
We introduce an inference technique to produce discriminative context-aware image captions (captions that describe differences between images or visual concepts) using only generic context-agnostic training data (captions that describe a concept or an image in isolation). For example, given images and captions of “siamese cat” and “tiger cat”, we generate language that describes the “siamese cat” in a way that distinguishes it from “tiger cat”. Our key novelty is that we show how to do joint inference over a language model that is context-agnostic and a listener which distinguishes closely-related concepts. We first apply our technique to a justification task, namely to describe why an image contains a particular fine-grained category as opposed to another closely-related category of the CUB-200-2011 dataset. We then study discriminative image captioning to generate language that uniquely refers to one of two semantically-similar images in the COCO dataset. Evaluations with discriminative ground truth for justification and human studies for discriminative image captioning reveal that our approach outperforms baseline generative and speaker-listener approaches for discrimination.
https://arxiv.org/abs/1701.02870
We investigate techniques for supervised domain adaptation for neural machine translation where an existing model trained on a large out-of-domain dataset is adapted to a small in-domain dataset. In this scenario, overfitting is a major challenge. We investigate a number of techniques to reduce overfitting and improve transfer learning, including regularization techniques such as dropout and L2-regularization towards an out-of-domain prior. In addition, we introduce tuneout, a novel regularization technique inspired by dropout. We apply these techniques, alone and in combination, to neural machine translation, obtaining improvements on IWSLT datasets for English->German and English->Russian. We also investigate the amounts of in-domain training data needed for domain adaptation in NMT, and find a logarithmic relationship between the amount of training data and gain in BLEU score.
https://arxiv.org/abs/1707.09920
Hashing has been widely used for large-scale search due to its low storage cost and fast query speed. By using supervised information, supervised hashing can significantly outperform unsupervised hashing. Recently, discrete supervised hashing and deep hashing are two representative progresses in supervised hashing. On one hand, hashing is essentially a discrete optimization problem. Hence, utilizing supervised information to directly guide discrete (binary) coding procedure can avoid sub-optimal solution and improve the accuracy. On the other hand, deep hashing, which integrates deep feature learning and hash-code learning into an end-to-end architecture, can enhance the feedback between feature learning and hash-code learning. The key in discrete supervised hashing is to adopt supervised information to directly guide the discrete coding procedure in hashing. The key in deep hashing is to adopt the supervised information to directly guide the deep feature learning procedure. However, there have not existed works which can use the supervised information to directly guide both discrete coding procedure and deep feature learning procedure in the same framework. In this paper, we propose a novel deep hashing method, called deep discrete supervised hashing (DDSH), to address this problem. DDSH is the first deep hashing method which can utilize supervised information to directly guide both discrete coding procedure and deep feature learning procedure, and thus enhance the feedback between these two important procedures. Experiments on three real datasets show that DDSH can outperform other state-of-the-art baselines, including both discrete hashing and deep hashing baselines, for image retrieval.
https://arxiv.org/abs/1707.09905
The necessity of using a fixed-size word vocabulary in order to control the model complexity in state-of-the-art neural machine translation (NMT) systems is an important bottleneck on performance, especially for morphologically rich languages. Conventional methods that aim to overcome this problem by using sub-word or character-level representations solely rely on statistics and disregard the linguistic properties of words, which leads to interruptions in the word structure and causes semantic and syntactic losses. In this paper, we propose a new vocabulary reduction method for NMT, which can reduce the vocabulary of a given input corpus at any rate while also considering the morphological properties of the language. Our method is based on unsupervised morphology learning and can be, in principle, used for pre-processing any language pair. We also present an alternative word segmentation method based on supervised morphological analysis, which aids us in measuring the accuracy of our model. We evaluate our method in Turkish-to-English NMT task where the input language is morphologically rich and agglutinative. We analyze different representation methods in terms of translation accuracy as well as the semantic and syntactic properties of the generated output. Our method obtains a significant improvement of 2.3 BLEU points over the conventional vocabulary reduction technique, showing that it can provide better accuracy in open vocabulary translation of morphologically rich languages.
https://arxiv.org/abs/1707.09879
Deep convolutional neural networks follow roughly the architecture of biological visual systems, and have shown a performance comparable to human observers in object recognition tasks. In this study, I test a pre-trained deep neural network in some classic visual search tasks. The results reveal a qualitative difference from human performance. It appears that there is no difference between searches for simple features that pop out in experiments with humans, and for feature configurations that exhibit strict capacity limitations in human vision. Both types of stimuli reveal moderate capacity limitations in the neural network tested here.
https://arxiv.org/abs/1707.09775
Positron emission tomography (PET) image synthesis plays an important role, which can be used to boost the training data for computer aided diagnosis systems. However, existing image synthesis methods have problems in synthesizing the low resolution PET images. To address these limitations, we propose multi-channel generative adversarial networks (M-GAN) based PET image synthesis method. Different to the existing methods which rely on using low-level features, the proposed M-GAN is capable to represent the features in a high-level of semantic based on the adversarial learning concept. In addition, M-GAN enables to take the input from the annotation (label) to synthesize the high uptake regions e.g., tumors and from the computed tomography (CT) images to constrain the appearance consistency and output the synthetic PET images directly. Our results on 50 lung cancer PET-CT studies indicate that our method was much closer to the real PET images when compared with the existing methods.
https://arxiv.org/abs/1707.09747
This paper compares the efficiency of various algorithms for implementing quantum resistant public key encryption scheme RLCE on 64-bit CPUs. By optimizing various algorithms for polynomial and matrix operations over finite fields, we obtained several interesting (or even surprising) results. For example, it is well known (e.g., Moenck 1976 \cite{moenck1976practical}) that Karatsuba’s algorithm outperforms classical polynomial multiplication algorithm from the degree 15 and above (practically, Karatsuba’s algorithm only outperforms classical polynomial multiplication algorithm from the degree 35 and above ). Our experiments show that 64-bit optimized Karatsuba’s algorithm will only outperform 64-bit optimized classical polynomial multiplication algorithm for polynomials of degree 115 and above over finite field $GF(2^{10})$. The second interesting (surprising) result shows that 64-bit optimized Chien’s search algorithm ourperforms all other 64-bit optimized polynomial root finding algorithms such as BTA and FFT for polynomials of all degrees over finite field $GF(2^{10})$. The third interesting (surprising) result shows that 64-bit optimized Strassen matrix multiplication algorithm only outperforms 64-bit optimized classical matrix multiplication algorithm for matrices of dimension 750 and above over finite field $GF(2^{10})$. It should be noted that existing literatures and practices recommend Strassen matrix multiplication algorithm for matrices of dimension 40 and above. All our experiments are done on a 64-bit MacBook Pro with i7 CPU and single thread C codes. It should be noted that the reported results should be appliable to 64 or larger bits CPU architectures. For 32 or smaller bits CPUs, these results may not be applicable. The source code and library for the algorithms covered in this paper are available at this http URL.
https://arxiv.org/abs/1702.07737
One central mystery of neural NLP is what neural models “know” about their subject matter. When a neural machine translation system learns to translate from one language to another, does it learn the syntax or semantics of the languages? Can this knowledge be extracted from the system to fill holes in human scientific knowledge? Existing typological databases contain relatively full feature specifications for only a few hundred languages. Exploiting the existence of parallel texts in more than a thousand languages, we build a massive many-to-one neural machine translation (NMT) system from 1017 languages into English, and use this to predict information missing from typological databases. Experiments show that the proposed method is able to infer not only syntactic, but also phonological and phonetic inventory features, and improves over a baseline that has access to information about the languages’ geographic and phylogenetic neighbors.
https://arxiv.org/abs/1707.09569
Learning to hash has been widely applied to approximate nearest neighbor search for large-scale multimedia retrieval, due to its computation efficiency and retrieval quality. Deep learning to hash, which improves retrieval quality by end-to-end representation learning and hash encoding, has received increasing attention recently. Subject to the ill-posed gradient difficulty in the optimization with sign activations, existing deep learning to hash methods need to first learn continuous representations and then generate binary hash codes in a separated binarization step, which suffer from substantial loss of retrieval quality. This work presents HashNet, a novel deep architecture for deep learning to hash by continuation method with convergence guarantees, which learns exactly binary hash codes from imbalanced similarity data. The key idea is to attack the ill-posed gradient problem in optimizing deep networks with non-smooth binary activations by continuation method, in which we begin from learning an easier network with smoothed activation function and let it evolve during the training, until it eventually goes back to being the original, difficult to optimize, deep network with the sign activation function. Comprehensive empirical evidence shows that HashNet can generate exactly binary hash codes and yield state-of-the-art multimedia retrieval performance on standard benchmarks.
https://arxiv.org/abs/1702.00758
We examine the effects of particular orderings of sentence pairs on the on-line training of neural machine translation (NMT). We focus on two types of such orderings: (1) ensuring that each minibatch contains sentences similar in some aspect and (2) gradual inclusion of some sentence types as the training progresses (so called “curriculum learning”). In our English-to-Czech experiments, the internal homogeneity of minibatches has no effect on the training but some of our “curricula” achieve a small improvement over the baseline.
https://arxiv.org/abs/1707.09533
In recent years, the performance of object detection has advanced significantly with the evolving deep convolutional neural networks. However, the state-of-the-art object detection methods still rely on accurate bounding box annotations that require extensive human labelling. Object detection without bounding box annotations, i.e, weakly supervised detection methods, are still lagging far behind. As weakly supervised detection only uses image level labels and does not require the ground truth of bounding box location and label of each object in an image, it is generally very difficult to distill knowledge of the actual appearances of objects. Inspired by curriculum learning, this paper proposes an easy-to-hard knowledge transfer scheme that incorporates easy web images to provide prior knowledge of object appearance as a good starting point. While exploiting large-scale free web imagery, we introduce a sophisticated labour free method to construct a web dataset with good diversity in object appearance. After that, semantic relevance and distribution relevance are introduced and utilized in the proposed curriculum training scheme. Our end-to-end learning with the constructed web data achieves remarkable improvement across most object classes especially for the classes that are often considered hard in other works.
https://arxiv.org/abs/1707.08721
The Earth observation satellites have been monitoring the earth’s surface for a long time, and the images taken by the satellites contain large amounts of valuable data. However, it is extremely hard work to manually analyze such huge data. Thus, a method of automatic object detection is needed for satellite images to facilitate efficient data analyses. This paper describes a new image feature extended from higher-order local autocorrelation to the object detection of multispectral satellite images. The feature has been extended to extract spectral inter-relationships in addition to spatial relationships to fully exploit multispectral information. The results of experiments with object detection tasks conducted to evaluate the effectiveness of the proposed feature extension indicate that the feature realized a higher performance compared to existing methods.
https://arxiv.org/abs/1707.09099
Machine comprehension(MC) style question answering is a representative problem in natural language processing. Previous methods rarely spend time on the improvement of encoding layer, especially the embedding of syntactic information and name entity of the words, which are very crucial to the quality of encoding. Moreover, existing attention methods represent each query word as a vector or use a single vector to represent the whole query sentence, neither of them can handle the proper weight of the key words in query sentence. In this paper, we introduce a novel neural network architecture called Multi-layer Embedding with Memory Network(MEMEN) for machine reading task. In the encoding layer, we employ classic skip-gram model to the syntactic and semantic information of the words to train a new kind of embedding layer. We also propose a memory network of full-orientation matching of the query and passage to catch more pivotal information. Experiments show that our model has competitive results both from the perspectives of precision and efficiency in Stanford Question Answering Dataset(SQuAD) among all published results and achieves the state-of-the-art results on TriviaQA dataset.
https://arxiv.org/abs/1707.09098
In recent years, three-dimensional GaN-based transistors have been intensively studied for their dramatically improved output power, better gate controllability, and shorter channels for speedup and miniaturization. However, thermal analysis of such devices is often oversimplified using the conventional Fourier’s law and bulk material properties in thermal simulations. In this aspect, accurate temperature predictions can be achieved by coupled phonon and electron Monte Carlo simulations that track the movement and scattering of individual phonons and electrons. However, the heavy computational load often restricts such simulations to nanoscale devices, while a real chip is of millimeter to centimeter sizes. This issue can be addressed by a hybrid simulation technique that employs the Fourier’s law for regions away from the hot spot. Using this technique, accurate electrothermal simulations are carried out on a nanowire-based GaN transistor to reveal the temperature rise in such devices.
https://arxiv.org/abs/1707.09073
Small- and large-signal RF characteristics were measured on AlN GaN HEMTs with 80-160 nm gate length and 100-300 {\mu}m width. Consistent with the literature, current-gain cut-off frequency and maximum frequency of oscillation were found to increase with inverse gate length and independent of gate width. For the first time, output power and efficiency were reported at the high end of Xband, and were comparable to the best reported at 2 GHz and insensitive to gate length or width. These results suggest that the AlN/GaN HEMTs can be further scaled for even higher frequency and higher power performance.
https://arxiv.org/abs/1707.09065
Explaining and reasoning about processes which underlie observed black-box phenomena enables the discovery of causal mechanisms, derivation of suitable abstract representations and the formulation of more robust predictions. We propose to learn high level functional programs in order to represent abstract models which capture the invariant structure in the observed data. We introduce the $\pi$-machine (program-induction machine) – an architecture able to induce interpretable LISP-like programs from observed data traces. We propose an optimisation procedure for program learning based on backpropagation, gradient descent and A* search. We apply the proposed method to two problems: system identification of dynamical systems and explaining the behaviour of a DQN agent. Our results show that the $\pi$-machine can efficiently induce interpretable programs from individual data traces.
https://arxiv.org/abs/1708.00376
With recent innovations in dense image captioning, it is now possible to describe every object of the scene with a caption while objects are determined by bounding boxes. However, interpretation of such an output is not trivial due to the existence of many overlapping bounding boxes. Furthermore, in current captioning frameworks, the user is not able to involve personal preferences to exclude out of interest areas. In this paper, we propose a novel hybrid deep learning architecture for interactive region segmentation and captioning where the user is able to specify an arbitrary region of the image that should be processed. To this end, a dedicated Fully Convolutional Network (FCN) named Lyncean FCN (LFCN) is trained using our special training data to isolate the User Intention Region (UIR) as the output of an efficient segmentation. In parallel, a dense image captioning model is utilized to provide a wide variety of captions for that region. Then, the UIR will be explained with the caption of the best match bounding box. To the best of our knowledge, this is the first work that provides such a comprehensive output. Our experiments show the superiority of the proposed approach over state-of-the-art interactive segmentation methods on several well-known datasets. In addition, replacement of the bounding boxes with the result of the interactive segmentation leads to a better understanding of the dense image captioning output as well as accuracy enhancement for the object detection in terms of Intersection over Union (IoU).
https://arxiv.org/abs/1707.08364
By monitoring a large number of stars in the Local Group galaxies, we can detect nanolensing events by sub-lunar mass compact objects (SULCOs) such as primordial black holes (PBHs) and rogue (free-floating) dwarf planets in the Milky Way halo. In contarst to microlensing by stellar-mass objects, the finite-source size effect becomes important and the lensing time duration becomes shorter ($\sim 10^{1-4}\,\textrm{s}$). Using stars with $V<26$ in M33 as sources, for one-night observation, we would be able to detect $10^{3-4}$ nanolensing events caused by SULCOs in the Milky Way halo with a mass of $10^{-9}M_{\odot}$ to $10^{-7}M_{\odot}$ for sources with S/N$>5$ if SULCOs constitute all the dark matter components. Moreover, we expect $10^{1-2}$ events in which bright blue stars with S/N$>100$ are weakly amplified due to lensing by SULCOs with a mass range of $10^{-11}M_{\odot}$ to $10^{-9}M_{\odot}$. Thus the method would open a new window on SULCOs in the Milky Way halo that would otherwise not be observable.
https://arxiv.org/abs/1609.04520
We explore different approaches to integrating a simple convolutional neural network (CNN) with the Lucene search engine in a multi-stage ranking architecture. Our models are trained using the PyTorch deep learning toolkit, which is implemented in C/C++ with a Python frontend. One obvious integration strategy is to expose the neural network directly as a service. For this, we use Apache Thrift, a software framework for building scalable cross-language services. In exploring alternative architectures, we observe that once trained, the feedforward evaluation of neural networks is quite straightforward. Therefore, we can extract the parameters of a trained CNN from PyTorch and import the model into Java, taking advantage of the Java Deeplearning4J library for feedforward evaluation. This has the advantage that the entire end-to-end system can be implemented in Java. As a third approach, we can extract the neural network from PyTorch and “compile” it into a C++ program that exposes a Thrift service. We evaluate these alternatives in terms of performance (latency and throughput) as well as ease of integration. Experiments show that feedforward evaluation of the convolutional neural network is significantly slower in Java, while the performance of the compiled C++ network does not consistently beat the PyTorch implementation.
https://arxiv.org/abs/1707.08275
This paper presents a state-of-the-art approach in object detection for being applied in future SLAM problems. Although, many SLAM methods are proposed to create suitable autonomy for mobile robots namely ground vehicles, they still face overconfidence and large computations during entrance to immense spaces with many landmarks. In particular, they suffer from impractical applications via sole reliance on the limited sensors like camera. Proposed method claims that unmanned ground vehicles without having huge amount of database for object definition and highly advance prediction parameters can deal with incoming objects during straight motion of camera in real-time. Line-Circle (LC) filter tries to apply detection, tracking and learning to each defined experts to obtain more information for judging scene without over-calculation. In this filter, circle expert let us summarize edges in groups. The Interactive feedback learning between each expert creates minimal error that fights against overwhelming landmark signs in crowded scenes without mapping. Our experts basically are dependent on trust factors’ covariance with geometric definitions to ignore, emerge and compare detected landmarks. The experiment for validating the model is taken place utilizing a camera beside an IMU sensor for location estimation.
https://arxiv.org/abs/1707.08095
We describe our recently developed neural machine translation (NMT) system and benchmark it against our own statistical machine translation (SMT) system as well as two other general purpose online engines (statistical and neural). We present automatic and human evaluation results of the translation output provided by each system. We also analyze the effect of sentence length on the quality of output for SMT and NMT systems.
https://arxiv.org/abs/1707.07911
In this paper, we address the task of detecting semantic parts on partially occluded objects. We consider a scenario where the model is trained using non-occluded images but tested on occluded images. The motivation is that there are infinite number of occlusion patterns in real world, which cannot be fully covered in the training data. So the models should be inherently robust and adaptive to occlusions instead of fitting / learning the occlusion patterns in the training data. Our approach detects semantic parts by accumulating the confidence of local visual cues. Specifically, the method uses a simple voting method, based on log-likelihood ratio tests and spatial constraints, to combine the evidence of local cues. These cues are called visual concepts, which are derived by clustering the internal states of deep networks. We evaluate our voting scheme on the VehicleSemanticPart dataset with dense part annotations. We randomly place two, three or four irrelevant objects onto the target object to generate testing images with various occlusions. Experiments show that our algorithm outperforms several competitors in semantic part detection when occlusions are present.
https://arxiv.org/abs/1707.07819
Unsupervised learning in a generalized Hopfield associative-memory network is investigated in this work. First, we prove that the (generalized) Hopfield model is equivalent to a semi-restricted Boltzmann machine with a layer of visible neurons and another layer of hidden binary neurons, so it could serve as the building block for a multilayered deep-learning system. We then demonstrate that the Hopfield network can learn to form a faithful internal representation of the observed samples, with the learned memory patterns being prototypes of the input data. Furthermore, we propose a spectral method to extract a small set of concepts (idealized prototypes) as the most concise summary or abstraction of the empirical data.
https://arxiv.org/abs/1704.02848
Time is an important relevance signal when searching streams of social media posts. The distribution of document timestamps from the results of an initial query can be leveraged to infer the distribution of relevant documents, which can then be used to rerank the initial results. Previous experiments have shown that kernel density estimation is a simple yet effective implementation of this idea. This paper explores an alternative approach to mining temporal signals with recurrent neural networks. Our intuition is that neural networks provide a more expressive framework to capture the temporal coherence of neighboring documents in time. To our knowledge, we are the first to integrate lexical and temporal signals in an end-to-end neural network architecture, in which existing neural ranking models are used to generate query-document similarity vectors that feed into a bidirectional LSTM layer for temporal modeling. Our results are mixed: existing neural models for document ranking alone yield limited improvements over simple baselines, but the integration of lexical and temporal signals yield significant improvements over competitive temporal baselines.
https://arxiv.org/abs/1707.07792
We make distributed stochastic gradient descent faster by exchanging sparse updates instead of dense updates. Gradient updates are positively skewed as most updates are near zero, so we map the 99% smallest updates (by absolute value) to zero then exchange sparse matrices. This method can be combined with quantization to further improve the compression. We explore different configurations and apply them to neural machine translation and MNIST image classification tasks. Most configurations work on MNIST, whereas different configurations reduce convergence rate on the more complex translation task. Our experiments show that we can achieve up to 49% speed up on MNIST and 22% on NMT without damaging the final accuracy or BLEU.
https://arxiv.org/abs/1704.05021
We provide the first observation of weak localization in high carrier density two-dimensional electron gas in AlInN/GaN heterostructures; at low temperatures and low fields the conductivity increases with increasing magnetic field. Weak localization is further confirmed by the lnT dependence of the zero-field conductivity and angle dependence of magnetoresistance. The inelastic scattering rate is linearly proportional to temperature, demonstrating that electron-electron scattering is the principal phase breaking mechanism. Shubnikov-de Haas (SdH) oscillations at high magnetic fields are also observed. From the temperature dependent amplitude of SdH oscillation and Dingle plot, the effective mass of electron is extracted to be 0.2327me; in addition the quantum lifetime is smaller than transport time from Hall measurement, indicating small angle scattering such as from remote ionized impurities is dominant. Above 20 K, the scattering changes from acoustic phonon to optical phonon scattering, resulting in a rapid decrease in carrier mobility with increasing temperature.
https://arxiv.org/abs/1707.07722
GaN is a key material for lighting technology. Yet, the carrier transport and ultrafast dynamics that are central in GaN light emitting devices are not completely understood. We present first-principles calculations of carrier dynamics in GaN, focusing on electron-phonon (e-ph) scattering and the cooling and nanoscale dynamics of hot carriers. We find that e-ph scattering is significantly faster for holes compared to electrons, and that for hot carriers with an initial 0.5$-$1 eV excess energy, holes take a significantly shorter time ($\sim$0.1 ps) to relax to the band edge compared to electrons, which take $\sim$1 ps. The asymmetry in the hot carrier dynamics is shown to originate from the valence band degeneracy, the heavier effective mass of holes compared to electrons, and the details of the coupling to different phonon modes in the valence and conduction bands. We show that the slow cooling of hot electrons and their long ballistic mean free paths (over 3 nm) are a possible cause of efficiency droop in GaN light emitting diodes. Taken together, our work sheds light on the ultrafast dynamics of hot carriers in GaN and the nanoscale origin of efficiency droop.
https://arxiv.org/abs/1703.07880
This paper presents a novel neural machine translation model which jointly learns translation and source-side latent graph representations of sentences. Unlike existing pipelined approaches using syntactic parsers, our end-to-end model learns a latent graph parser as part of the encoder of an attention-based neural machine translation model, and thus the parser is optimized according to the translation objective. In experiments, we first show that our model compares favorably with state-of-the-art sequential and pipelined syntax-based NMT models. We also show that the performance of our model can be further improved by pre-training it with a small amount of treebank annotations. Our final ensemble model significantly outperforms the previous best models on the standard English-to-Japanese translation dataset.
https://arxiv.org/abs/1702.02265
In this work, we perform an empirical comparison among the CTC, RNN-Transducer, and attention-based Seq2Seq models for end-to-end speech recognition. We show that, without any language model, Seq2Seq and RNN-Transducer models both outperform the best reported CTC models with a language model, on the popular Hub5’00 benchmark. On our internal diverse dataset, these trends continue - RNNTransducer models rescored with a language model after beam search outperform our best CTC models. These results simplify the speech recognition pipeline so that decoding can now be expressed purely as neural network operations. We also study how the choice of encoder architecture affects the performance of the three models - when all encoder layers are forward only, and when encoders downsample the input representation aggressively.
https://arxiv.org/abs/1707.07413
Often multiple instances of an object occur in the same scene, for example in a warehouse. Unsupervised multi-instance object discovery algorithms are able to detect and identify such objects. We use such an algorithm to provide object proposals to a convolutional neural network (CNN) based classifier. This results in fewer regions to evaluate, compared to traditional region proposal algorithms. Additionally, it enables using the joint probability of multiple instances of an object, resulting in improved classification accuracy. The proposed technique can also split a single class into multiple sub-classes corresponding to the different object types, enabling hierarchical classification.
https://arxiv.org/abs/1707.07255
In translation, considering the document as a whole can help to resolve ambiguities and inconsistencies. In this paper, we propose a cross-sentence context-aware approach and investigate the influence of historical contextual information on the performance of neural machine translation (NMT). First, this history is summarized in a hierarchical way. We then integrate the historical representation into NMT in two strategies: 1) a warm-start of encoder and decoder states, and 2) an auxiliary context source for updating decoder states. Experimental results on a large Chinese-English translation task show that our approach significantly improves upon a strong attention-based NMT system by up to +2.1 BLEU points.
https://arxiv.org/abs/1704.04347
We propose a novel decoding approach for neural machine translation (NMT) based on continuous optimisation. We convert decoding - basically a discrete optimization problem - into a continuous optimization problem. The resulting constrained continuous optimisation problem is then tackled using gradient-based methods. Our powerful decoding framework enables decoding intractable models such as the intersection of left-to-right and right-to-left (bidirectional) as well as source-to-target and target-to-source (bilingual) NMT models. Our empirical results show that our decoding framework is effective, and leads to substantial improvements in translations generated from the intersected models where the typical greedy or beam search is not feasible. We also compare our framework against reranking, and analyse its advantages and disadvantages.
https://arxiv.org/abs/1701.02854
Generating captions for images is a task that has recently received considerable attention. In this work we focus on caption generation for abstract scenes, or object layouts where the only information provided is a set of objects and their locations. We propose OBJ2TEXT, a sequence-to-sequence model that encodes a set of objects and their locations as an input sequence using an LSTM network, and decodes this representation using an LSTM language model. We show that our model, despite encoding object layouts as a sequence, can represent spatial relationships between objects, and generate descriptions that are globally coherent and semantically relevant. We test our approach in a task of object-layout captioning by using only object annotations as inputs. We additionally show that our model, combined with a state-of-the-art object detector, improves an image captioning model from 0.863 to 0.950 (CIDEr score) in the test benchmark of the standard MS-COCO Captioning task.
https://arxiv.org/abs/1707.07102
It has been well demonstrated that adversarial examples, i.e., natural images with visually imperceptible perturbations added, generally exist for deep networks to fail on image classification. In this paper, we extend adversarial examples to semantic segmentation and object detection which are much more difficult. Our observation is that both segmentation and detection are based on classifying multiple targets on an image (e.g., the basic target is a pixel or a receptive field in segmentation, and an object proposal in detection), which inspires us to optimize a loss function over a set of pixels/proposals for generating adversarial perturbations. Based on this idea, we propose a novel algorithm named Dense Adversary Generation (DAG), which generates a large family of adversarial examples, and applies to a wide range of state-of-the-art deep networks for segmentation and detection. We also find that the adversarial perturbations can be transferred across networks with different training data, based on different architectures, and even for different recognition tasks. In particular, the transferability across networks with the same architecture is more significant than in other cases. Besides, summing up heterogeneous perturbations often leads to better transfer performance, which provides an effective method of black-box adversarial attack.
https://arxiv.org/abs/1703.08603
This paper introduces SGNMT, our experimental platform for machine translation research. SGNMT provides a generic interface to neural and symbolic scoring modules (predictors) with left-to-right semantic such as translation models like NMT, language models, translation lattices, $n$-best lists or other kinds of scores and constraints. Predictors can be combined with other predictors to form complex decoding tasks. SGNMT implements a number of search strategies for traversing the space spanned by the predictors which are appropriate for different predictor constellations. Adding new predictors or decoding strategies is particularly easy, making it a very efficient tool for prototyping new research ideas. SGNMT is actively being used by students in the MPhil program in Machine Learning, Speech and Language Technology at the University of Cambridge for course work and theses, as well as for most of the research work in our group.
https://arxiv.org/abs/1707.06885
Ensembling is a well-known technique in neural machine translation (NMT) to improve system performance. Instead of a single neural net, multiple neural nets with the same topology are trained separately, and the decoder generates predictions by averaging over the individual models. Ensembling often improves the quality of the generated translations drastically. However, it is not suitable for production systems because it is cumbersome and slow. This work aims to reduce the runtime to be on par with a single system without compromising the translation quality. First, we show that the ensemble can be unfolded into a single large neural network which imitates the output of the ensemble system. We show that unfolding can already improve the runtime in practice since more work can be done on the GPU. We proceed by describing a set of techniques to shrink the unfolded network by reducing the dimensionality of layers. On Japanese-English we report that the resulting network has the size and decoding speed of a single NMT network but performs on the level of a 3-ensemble system.
https://arxiv.org/abs/1704.03279
We define the object detection from imagery problem as estimating a very large but extremely sparse bounding box dependent probability distribution. Subsequently we identify a sparse distribution estimation scheme, Directed Sparse Sampling, and employ it in a single end-to-end CNN based detection model. This methodology extends and formalizes previous state-of-the-art detection models with an additional emphasis on high evaluation rates and reduced manual engineering. We introduce two novelties, a corner based region-of-interest estimator and a deconvolution based CNN model. The resulting model is scene adaptive, does not require manually defined reference bounding boxes and produces highly competitive results on MSCOCO, Pascal VOC 2007 and Pascal VOC 2012 with real-time evaluation rates. Further analysis suggests our model performs particularly well when finegrained object localization is desirable. We argue that this advantage stems from the significantly larger set of available regions-of-interest relative to other methods. Source-code is available from: this https URL
https://arxiv.org/abs/1703.10295
Synthesizing high resolution photorealistic images has been a long-standing challenge in machine learning. In this paper we introduce new methods for the improved training of generative adversarial networks (GANs) for image synthesis. We construct a variant of GANs employing label conditioning that results in 128x128 resolution image samples exhibiting global coherence. We expand on previous work for image quality assessment to provide two new analyses for assessing the discriminability and diversity of samples from class-conditional image synthesis models. These analyses demonstrate that high resolution samples provide class information not present in low resolution samples. Across 1000 ImageNet classes, 128x128 samples are more than twice as discriminable as artificially resized 32x32 samples. In addition, 84.7% of the classes have samples exhibiting diversity comparable to real ImageNet data.
https://arxiv.org/abs/1610.09585
Recent captioning models are limited in their ability to scale and describe concepts unseen in paired image-text corpora. We propose the Novel Object Captioner (NOC), a deep visual semantic captioning model that can describe a large number of object categories not present in existing image-caption datasets. Our model takes advantage of external sources – labeled images from object recognition datasets, and semantic knowledge extracted from unannotated text. We propose minimizing a joint objective which can learn from these diverse data sources and leverage distributional semantic embeddings, enabling the model to generalize and describe novel objects outside of image-caption datasets. We demonstrate that our model exploits semantic information to generate captions for hundreds of object categories in the ImageNet object recognition dataset that are not observed in MSCOCO image-caption training data, as well as many categories that are observed very rarely. Both automatic evaluations and human judgements show that our model considerably outperforms prior work in being able to describe many more categories of objects.
https://arxiv.org/abs/1606.07770
The Breakthrough Listen Initiative is undertaking a comprehensive search for radio and optical signatures from extraterrestrial civilizations. An integral component of the project is the design and implementation of wide-bandwidth data recorder and signal processing systems. The capabilities of these systems, particularly at radio frequencies, directly determine survey speed; further, given a fixed observing time and spectral coverage, they determine sensitivity as well. Here, we detail the Breakthrough Listen wide-bandwidth data recording system deployed at the 100-m aperture Robert C. Byrd Green Bank Telescope. The system digitizes up to 6 GHz of bandwidth at 8 bits for both polarizations, storing the resultant 24 GB/s of data to disk. This system is among the highest data rate baseband recording systems in use in radio astronomy. A future system expansion will double recording capacity, to achieve a total Nyquist bandwidth of 12 GHz in two polarizations. In this paper, we present details of the system architecture, along with salient configuration and disk-write optimizations used to achieve high-throughput data capture on commodity compute servers and consumer-class hard disk drives.
https://arxiv.org/abs/1707.06024
Object detection aims at high speed and accuracy simultaneously. However, fast models are usually less accurate, while accurate models cannot satisfy our need for speed. A fast model can be 10 times faster but 50\% less accurate than an accurate model. In this paper, we propose Adaptive Feeding (AF) to combine a fast (but less accurate) detector and an accurate (but slow) detector, by adaptively determining whether an image is easy or hard and choosing an appropriate detector for it. In practice, we build a cascade of detectors, including the AF classifier which make the easy vs. hard decision and the two detectors. The AF classifier can be tuned to obtain different tradeoff between speed and accuracy, which has negligible training time and requires no additional training data. Experimental results on the PASCAL VOC, MS COCO and Caltech Pedestrian datasets confirm that AF has the ability to achieve comparable speed as the fast detector and comparable accuracy as the accurate one at the same time. As an example, by combining the fast SSD300 with the accurate SSD500 detector, AF leads to 50\% speedup over SSD500 with the same precision on the VOC2007 test set.
https://arxiv.org/abs/1707.06399
Existing image captioning models do not generalize well to out-of-domain images containing novel scenes or objects. This limitation severely hinders the use of these models in real world applications dealing with images in the wild. We address this problem using a flexible approach that enables existing deep captioning architectures to take advantage of image taggers at test time, without re-training. Our method uses constrained beam search to force the inclusion of selected tag words in the output, and fixed, pretrained word embeddings to facilitate vocabulary expansion to previously unseen tag words. Using this approach we achieve state of the art results for out-of-domain captioning on MSCOCO (and improved results for in-domain captioning). Perhaps surprisingly, our results significantly outperform approaches that incorporate the same tag predictions into the learning algorithm. We also show that we can significantly improve the quality of generated ImageNet captions by leveraging ground-truth labels.
https://arxiv.org/abs/1612.00576
This paper is first-line research expanding GANs into graph topology analysis. By leveraging the hierarchical connectivity structure of a graph, we have demonstrated that generative adversarial networks (GANs) can successfully capture topological features of any arbitrary graph, and rank edge sets by different stages according to their contribution to topology reconstruction. Moreover, in addition to acting as an indicator of graph reconstruction, we find that these stages can also preserve important topological features in a graph.
https://arxiv.org/abs/1707.06197
Existing region-based object detectors are limited to regions with fixed box geometry to represent objects, even if those are highly non-rectangular. In this paper we introduce DP-FCN, a deep model for object detection which explicitly adapts to shapes of objects with deformable parts. Without additional annotations, it learns to focus on discriminative elements and to align them, and simultaneously brings more invariance for classification and geometric information to refine localization. DP-FCN is composed of three main modules: a Fully Convolutional Network to efficiently maintain spatial resolution, a deformable part-based RoI pooling layer to optimize positions of parts and build invariance, and a deformation-aware localization module explicitly exploiting displacements of parts to improve accuracy of bounding box regression. We experimentally validate our model and show significant gains. DP-FCN achieves state-of-the-art performances of 83.1% and 80.9% on PASCAL VOC 2007 and 2012 with VOC data only.
https://arxiv.org/abs/1707.06175
The attention mechanisms in deep neural networks are inspired by human’s attention that sequentially focuses on the most relevant parts of the information over time to generate prediction output. The attention parameters in those models are implicitly trained in an end-to-end manner, yet there have been few trials to explicitly incorporate human gaze tracking to supervise the attention models. In this paper, we investigate whether attention models can benefit from explicit human gaze labels, especially for the task of video captioning. We collect a new dataset called VAS, consisting of movie clips, and corresponding multiple descriptive sentences along with human gaze tracking data. We propose a video captioning model named Gaze Encoding Attention Network (GEAN) that can leverage gaze tracking information to provide the spatial and temporal attention for sentence generation. Through evaluation of language similarity metrics and human assessment via Amazon mechanical Turk, we demonstrate that spatial attentions guided by human gaze data indeed improve the performance of multiple captioning methods. Moreover, we show that the proposed approach achieves the state-of-the-art performance for both gaze prediction and video captioning not only in our VAS dataset but also in standard datasets (e.g. LSMDC and Hollywood2).
https://arxiv.org/abs/1707.06029
State-of-the-art methods for 3D hand pose estimation from depth images require large amounts of annotated training data. We propose to model the statistical relationships of 3D hand poses and corresponding depth images using two deep generative models with a shared latent space. By design, our architecture allows for learning from unlabeled image data in a semi-supervised manner. Assuming a one-to-one mapping between a pose and a depth map, any given point in the shared latent space can be projected into both a hand pose and a corresponding depth map. Regressing the hand pose can then be done by learning a discriminator to estimate the posterior of the latent pose given some depth maps. To improve generalization and to better exploit unlabeled depth maps, we jointly train a generator and a discriminator. At each iteration, the generator is updated with the back-propagated gradient from the discriminator to synthesize realistic depth maps of the articulated hand, while the discriminator benefits from an augmented training set of synthesized and unlabeled samples. The proposed discriminator network architecture is highly efficient and runs at 90 FPS on the CPU with accuracies comparable or better than state-of-art on 3 publicly available benchmarks.
https://arxiv.org/abs/1702.03431
Object detection is an essential task for autonomous robots operating in dynamic and changing environments. A robot should be able to detect objects in the presence of sensor noise that can be induced by changing lighting conditions for cameras and false depth readings for range sensors, especially RGB-D cameras. To tackle these challenges, we propose a novel adaptive fusion approach for object detection that learns weighting the predictions of different sensor modalities in an online manner. Our approach is based on a mixture of convolutional neural network (CNN) experts and incorporates multiple modalities including appearance, depth and motion. We test our method in extensive robot experiments, in which we detect people in a combined indoor and outdoor scenario from RGB-D data, and we demonstrate that our method can adapt to harsh lighting changes and severe camera motion blur. Furthermore, we present a new RGB-D dataset for people detection in mixed in- and outdoor environments, recorded with a mobile robot.
https://arxiv.org/abs/1707.05733
Neural machine translation (NMT) models are able to partially learn syntactic information from sequential lexical information. Still, some complex syntactic phenomena such as prepositional phrase attachment are poorly modeled. This work aims to answer two questions: 1) Does explicitly modeling target language syntax help NMT? 2) Is tight integration of words and syntax better than multitask training? We introduce syntactic information in the form of CCG supertags in the decoder, by interleaving the target supertags with the word sequence. Our results on WMT data show that explicitly modeling target-syntax improves machine translation quality for German->English, a high-resource pair, and for Romanian->English, a low-resource pair and also several syntactic phenomena including prepositional phrase attachment. Furthermore, a tight coupling of words and syntax improves translation quality more than multitask training. By combining target-syntax with adding source-side dependency labels in the embedding layer, we obtain a total improvement of 0.9 BLEU for German->English and 1.2 BLEU for Romanian->English.
https://arxiv.org/abs/1702.01147