Searcharxiv⌕ Search

arXiv subjects

Kun He

Publications and source records attributed to Kun He.

At least 199 records · Page 11Linked to original sources
arXiv2018

Child Gender Determination with Convolutional Neural Networks on Hand Radio-Graphs

Motivation: In forensic or medico-legal investigation as well as in anthropology the gender determination of the subject (hit by a disastrous or any kind of traumatic situation) is mostly the first step. In state-of-the-art techniques the gender is determined by examining dimensions of the bones of skull and the pelvis area. In worse situations when there is only a small portion of the human remains to be investigated and the subject is a child, we need alternative techniques to determine the gender of the subject. In this work we propose a technique called GDCNN (Gender Determination with Convolutional Neural Networks), where the left hand radio-graphs of the children between a wide range of ages in 1 month to 18 years are examined to determine the gender. To our knowledge this technique is first of its kind. Further to identify the area of the attention we used Class Activation Mapping (CAM). Results: The results suggest the accuracy of the model is as high as 98%, which is very convincing by taking into account the incompletely grown skeleton of the children. The attention observed with CAM discovers that the lower part of the hand around carpals (wrist) is more important for child gender determination.

cs.CV↗
arXiv2018

Hashing as Tie-Aware Learning to Rank

Hashing, or learning binary embeddings of data, is frequently used in nearest neighbor retrieval. In this paper, we develop learning to rank formulations for hashing, aimed at directly optimizing ranking-based evaluation metrics such as Average Precision (AP) and Normalized Discounted Cumulative Gain (NDCG). We first observe that the integer-valued Hamming distance often leads to tied rankings, and propose to use tie-aware versions of AP and NDCG to evaluate hashing for retrieval. Then, to optimize tie-aware ranking metrics, we derive their continuous relaxations, and perform gradient-based optimization with deep neural networks. Our results establish the new state-of-the-art for image retrieval by Hamming ranking in common benchmarks.

stat.ML↗
arXiv2018

An Iterative Path-Breaking Approach with Mutation and Restart Strategies for the MAX-SAT Problem

Although Path-Relinking is an effective local search method for many combinatorial optimization problems, its application is not straightforward in solving the MAX-SAT, an optimization variant of the satisfiability problem (SAT) that has many real-world applications and has gained more and more attention in academy and industry. Indeed, it was not used in any recent competitive MAX-SAT algorithms in our knowledge. In this paper, we propose a new local search algorithm called IPBMR for the MAX-SAT, that remedies the drawbacks of the Path-Relinking method by using a careful combination of three components: a new strategy named Path-Breaking to avoid unpromising regions of the search space when generating trajectories between two elite solutions; a weak and a strong mutation strategies, together with restarts, to diversify the search; and stochastic path generating steps to avoid premature local optimum solutions. We then present experimental results to show that IPBMR outperforms two of the best state-of-the-art MAX-SAT solvers, and an empirical investigation to identify and explain the effect of the three components in IPBMR.

cs.AI↗
arXiv2018

Hashing with Binary Matrix Pursuit

We propose theoretical and empirical improvements for two-stage hashing methods. We first provide a theoretical analysis on the quality of the binary codes and show that, under mild assumptions, a residual learning scheme can construct binary codes that fit any neighborhood structure with arbitrary accuracy. Secondly, we show that with high-capacity hash functions such as CNNs, binary code inference can be greatly simplified for many standard neighborhood definitions, yielding smaller optimization problems and more robust codes. Incorporating our findings, we propose a novel two-stage hashing method that significantly outperforms previous hashing studies on widely used image retrieval benchmarks.

cs.LG↗
arXiv2018

Hashing with Mutual Information

Binary vector embeddings enable fast nearest neighbor retrieval in large databases of high-dimensional objects, and play an important role in many practical applications, such as image and video retrieval. We study the problem of learning binary vector embeddings under a supervised setting, also known as hashing. We propose a novel supervised hashing method based on optimizing an information-theoretic quantity: mutual information. We show that optimizing mutual information can reduce ambiguity in the induced neighborhood structure in the learned Hamming space, which is essential in obtaining high retrieval performance. To this end, we optimize mutual information in deep neural networks with minibatch stochastic gradient descent, with a formulation that maximally and efficiently utilizes available supervision. Experiments on four image retrieval benchmarks, including ImageNet, confirm the effectiveness of our method in learning high-quality binary embeddings for nearest neighbor retrieval.

cs.CV↗
arXiv2018

Local Descriptors Optimized for Average Precision

Extraction of local feature descriptors is a vital stage in the solution pipelines for numerous computer vision tasks. Learning-based approaches improve performance in certain tasks, but still cannot replace handcrafted features in general. In this paper, we improve the learning of local feature descriptors by optimizing the performance of descriptor matching, which is a common stage that follows descriptor extraction in local feature based pipelines, and can be formulated as nearest neighbor retrieval. Specifically, we directly optimize a ranking-based retrieval performance metric, Average Precision, using deep neural networks. This general-purpose solution can also be viewed as a listwise learning to rank approach, which is advantageous compared to recent local ranking approaches. On standard benchmarks, descriptors learned with our formulation achieve state-of-the-art results in patch verification, patch retrieval, and image matching.

cs.CV↗
arXiv2018

Randomness in Deconvolutional Networks for Visual Representation

Toward a deeper understanding on the inner work of deep neural networks, we investigate CNN (convolutional neural network) using DCN (deconvolutional network) and randomization technique, and gain new insights for the intrinsic property of this network architecture. For the random representations of an untrained CNN, we train the corresponding DCN to reconstruct the input images. Compared with the image inversion on pre-trained CNN, our training converges faster and the yielding network exhibits higher quality for image reconstruction. It indicates there is rich information encoded in the random features; the pre-trained CNN may discard information irrelevant for classification and encode relevant features in a way favorable for classification but harder for reconstruction. We further explore the property of the overall random CNN-DCN architecture. Surprisingly, images can be inverted with satisfactory quality. Extensive empirical evidence as well as theoretical analysis are provided.

cs.CV↗
arXiv2017

The Local Dimension of Deep Manifold

Based on our observation that there exists a dramatic drop for the singular values of the fully connected layers or a single feature map of the convolutional layer, and that the dimension of the concatenated feature vector almost equals the summation of the dimension on each feature map, we propose a singular value decomposition (SVD) based approach to estimate the dimension of the deep manifolds for a typical convolutional neural network VGG19. We choose three categories from the ImageNet, namely Persian Cat, Container Ship and Volcano, and determine the local dimension of the deep manifolds of the deep layers through the tangent space of a target image. Through several augmentation methods, we found that the Gaussian noise method is closer to the intrinsic dimension, as by adding random noise to an image we are moving in an arbitrary dimension, and when the rank of the feature matrix of the augmented images does not increase we are very close to the local dimension of the manifold. We also estimate the dimension of the deep manifold based on the tangent space for each of the maxpooling layers. Our results show that the dimensions of different categories are close to each other and decline quickly along the convolutional layers and fully connected layers. Furthermore, we show that the dimensions decline quickly inside the Conv5 layer. Our work provides new insights for the intrinsic structure of deep neural networks and helps unveiling the inner organization of the black box of deep neural networks.

cs.CV↗
arXiv2017

Rectangle Transformation Problem

In this paper, we propose the rectangle transformation problem (RTP) and its variants. RTP asks for a transformation by a rectangle partition between two rectangles of the same area. We are interested in the minimum RTP which requires to minimize the partition size. We mainly focus on the strict rectangle transformation problem (SRTP) in which rotation is not allowed in transforming. We show that SRTP has no finite solution if the ratio of the two parallel side lengths of input rectangles is irrational. So we turn to its complement denoted by SIRTP, in which case all side lengths can be assumed integral. We give a polynomial time algorithm ALGSIRTP which gives a solution at most $q/p+O(\sqrt{p})$ to SIRTP$(p,q)$ ($q\geq p$), where $p$ and $q$ are two integer side lengths of input rectangles $p\times q$ and $q\times p$, and so ALGSIRTP is a $O(\sqrt{p})$-approximation algorithm for minimum SIRTP$(p,q)$. On the other hand, we show that there is not constant solution to SIRTP$(p,q)$ for all integers $p$ and $q$ ($q>p$) even though the ratio $q/p$ is within any constant range. We also raise a series of open questions for the research along this line.

cs.CG↗
arXiv2017

Variable Version Lovász Local Lemma: Beyond Shearer's Bound

A tight criterion under which the abstract version Lovász Local Lemma (abstract-LLL) holds was given by Shearer decades ago. However, little is known about that of the variable version LLL (variable-LLL) where events are generated by independent random variables, though this model of events is applicable to almost all applications of LLL. We introduce a necessary and sufficient criterion for variable-LLL, in terms of the probabilities of the events and the event-variable graph specifying the dependency among the events. Based on this new criterion, we obtain boundaries for two families of event-variable graphs, namely, cyclic and treelike bigraphs. These are the first two non-trivial cases where the variable-LLL boundary is fully determined. As a byproduct, we also provide a universal constructive method to find a set of events whose union has the maximum probability, given the probability vector and the event-variable graph. Though it is #P-hard in general to determine variable-LLL boundaries, we can to some extent decide whether a gap exists between a variable-LLL boundary and the corresponding abstract-LLL boundary. In particular, we show that the gap existence can be decided without solving Shearer's conditions or checking our variable-LLL criterion. Equipped with this powerful theorem, we show that there is no gap if the base graph of the event-variable graph is a tree, while gap appears if the base graph has an induced cycle of length at least 4. The problem is almost completely solved except when the base graph has only 3-cliques, in which case we also get partial solutions. A set of reduction rules are established that facilitate to infer gap existence of an event-variable graph from known ones. As an application, various event-variable graphs, in particular combinatorial ones, are shown to be gapful/gapless.

cs.DM↗
arXiv2017

MIHash: Online Hashing with Mutual Information

Learning-based hashing methods are widely used for nearest neighbor retrieval, and recently, online hashing methods have demonstrated good performance-complexity trade-offs by learning hash functions from streaming data. In this paper, we first address a key challenge for online hashing: the binary codes for indexed data must be recomputed to keep pace with updates to the hash functions. We propose an efficient quality measure for hash functions, based on an information-theoretic quantity, mutual information, and use it successfully as a criterion to eliminate unnecessary hash table updates. Next, we also show how to optimize the mutual information objective using stochastic gradient descent. We thus develop a novel hashing method, MIHash, that can be used in both online and batch settings. Experiments on image retrieval benchmarks (including a 2.5M image dataset) confirm the effectiveness of our formulation, both in reducing hash table recomputations and in learning high-quality hash functions.

cs.CV↗
arXiv2017

Predicting Foreground Object Ambiguity and Efficiently Crowdsourcing the Segmentation(s)

We propose the ambiguity problem for the foreground object segmentation task and motivate the importance of estimating and accounting for this ambiguity when designing vision systems. Specifically, we distinguish between images which lead multiple annotators to segment different foreground objects (ambiguous) versus minor inter-annotator differences of the same object. Taking images from eight widely used datasets, we crowdsource labeling the images as "ambiguous" or "not ambiguous" to segment in order to construct a new dataset we call STATIC. Using STATIC, we develop a system that automatically predicts which images are ambiguous. Experiments demonstrate the advantage of our prediction system over existing saliency-based methods on images from vision benchmarks and images taken by blind people who are trying to recognize objects in their environment. Finally, we introduce a crowdsourcing system to achieve cost savings for collecting the diversity of all valid "ground truth" foreground object segmentations by collecting extra segmentations only when ambiguity is expected. Experiments show our system eliminates up to 47% of human effort compared to existing crowdsourcing methods with no loss in capturing the diversity of ground truths.

cs.CV↗
arXiv2017

An Efficient Quasi-physical Quasi-human Algorithm for Packing Equal Circles in a Circular Container

This paper addresses the equal circle packing problem, and proposes an efficient Quasi-physical Quasi-human (QPQH) algorithm. QPQH is based on a modified Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm which we call the local BFGS and a new basin hopping strategy based on a Chinese idiom: alternate tension with relaxation. Starting from a random initial layout, we apply the local BFGS algorithm to reach a local minimum layout. The local BFGS algorithm fully utilizes the neighborhood information of each circle to considerably speed up the running time of the gradient descent process, and the efficiency is very apparent for large scale instances. When yielding a local minimum layout, the new basin-hopping strategy is to shrink the container size to different extent to generate several new layouts. Experimental results indicate that the new basin-hopping strategy is very efficient, especially for a type of layout with comparatively dense packing in the center and comparatively sparse packing around the boundary of the container. We test QPQH on the instances of n = 1,2,...,320, and obtain 66 new layouts which have smaller container sizes than the current best-known results reported in literature.

cs.CG↗
arXiv2017

Packing Unequal Circles into a Square Container by Partitioning Narrow Action Spaces and Circle Items

We address the NP-hard problem of finding a non-overlapping dense packing pattern for n Unequal Circle items in a two-dimensional Square Container (PUC-SC) such that the size of the container is minimized. Based on our previous work on an Action Space based Global Optimization (ASGO) that approximates each circle item as a square item to efficiently find the large unoccupied spaces, we propose an optimization algorithm based on the Partitioned Action Space and Partitioned Circle Items (PAS-PCI). The PAS is to partition the narrow action space on the long side to find two equal action spaces to fully utilize the unoccupied spaces. The PCI is to partition the circle items into four groups based on size for the basin hopping strategy. Experiments on two sets of benchmark instances show the effectiveness of the proposed method. In comparison with our previous algorithm ASGO on the 68 tested instances that ASGO published, PAS-PCI not only gains smaller containers in 64 instances and matches the other 4 but also runs faster in most instances. In comparison with the best record of the Packomania website on a total of 98 instances, PAS-PCI finds smaller containers on 82 and matches the other 16. Note that we updated 19 records for (47-48, 51-54, 57, 61-72) that had been kept unchanged since 2013.

cs.CG↗
arXiv2017

Hidden Community Detection in Social Networks

We introduce a new paradigm that is important for community detection in the realm of network analysis. Networks contain a set of strong, dominant communities, which interfere with the detection of weak, natural community structure. When most of the members of the weak communities also belong to stronger communities, they are extremely hard to be uncovered. We call the weak communities the hidden community structure. We present a novel approach called HICODE (HIdden COmmunity DEtection) that identifies the hidden community structure as well as the dominant community structure. By weakening the strength of the dominant structure, one can uncover the hidden structure beneath. Likewise, by reducing the strength of the hidden structure, one can more accurately identify the dominant structure. In this way, HICODE tackles both tasks simultaneously. Extensive experiments on real-world networks demonstrate that HICODE outperforms several state-of-the-art community detection methods in uncovering both the dominant and the hidden structure. In the Facebook university social networks, we find multiple non-redundant sets of communities that are strongly associated with residential hall, year of registration or career position of the faculties or students, while the state-of-the-art algorithms mainly locate the dominant ground truth category. In the Due to the difficulty of labeling all ground truth communities in real-world datasets, HICODE provides a promising approach to pinpoint the existing latent communities and uncover communities for which there is no ground truth. Finding this unknown structure is an extremely important community detection problem.

cs.SI↗
arXiv2016

A Tighter Relation between Sensitivity and Certificate Complexity

The sensitivity conjecture which claims that the sensitivity complexity is polynomially related to block sensitivity complexity, is one of the most important and challenging problem in decision tree complexity theory. Despite of a lot of efforts, the best known upper bound of block sensitivity, as well as the certificate complexity, are still exponential in terms of sensitivity: $bs(f)\leq C(f)\leq\max\{2^{s(f)-1}(s(f)-\frac{1}{3}),s(f)\}$. In this paper, we give a better upper bound of $bs(f)\leq C(f)\leq(\frac{8}{9} + o(1))s(f)2^{s(f) - 1}$. The proof is based on a deep investigation on the structure of the sensitivity graph. We also provide a tighter relationship between $C_0(f)$ and $s_0(f)$ for functions with $s_1(f)=2$.

cs.CC↗
arXiv2016

A Powerful Generative Model Using Random Weights for the Deep Image Representation

To what extent is the success of deep visualization due to the training? Could we do deep visualization using untrained, random weight networks? To address this issue, we explore new and powerful generative models for three popular deep visualization tasks using untrained, random weight convolutional neural networks. First we invert representations in feature spaces and reconstruct images from white noise inputs. The reconstruction quality is statistically higher than that of the same method applied on well trained networks with the same architecture. Next we synthesize textures using scaled correlations of representations in multiple layers and our results are almost indistinguishable with the original natural texture and the synthesized textures based on the trained network. Third, by recasting the content of an image in the style of various artworks, we create artistic images with high perceptual quality, highly competitive to the prior work of Gatys et al. on pretrained networks. To our knowledge this is the first demonstration of image representations using untrained deep neural networks. Our work provides a new and fascinating tool to study the representation of deep network architecture and sheds light on new understandings on deep visualization.

cs.CV↗