arXiv Science⌕ Search

arXiv subjects

Chen Sun

Publications and source records attributed to Chen Sun.

245 records · Page 14Linked to original sources
arXiv2016

ACD: Action Concept Discovery from Image-Sentence Corpora

Action classification in still images is an important task in computer vision. It is challenging as the appearances of ac- tions may vary depending on their context (e.g. associated objects). Manually labeling of context information would be time consuming and difficult to scale up. To address this challenge, we propose a method to automatically discover and cluster action concepts, and learn their classifiers from weakly supervised image-sentence corpora. It obtains candidate action concepts by extracting verb-object pairs from sentences and verifies their visualness with the associated images. Candidate action concepts are then clustered by using a multi-modal representation with image embeddings from deep convolutional networks and text embeddings from word2vec. More than one hundred human action concept classifiers are learned from the Flickr 30k dataset with no additional human effort and promising classification results are obtained. We further apply the AdaBoost algorithm to automatically select and combine relevant action concepts given an action query. Promising results have been shown on the PASCAL VOC 2012 action classification benchmark, which has zero overlap with Flickr30k.

cs.CV↗
arXiv2016

ProNet: Learning to Propose Object-specific Boxes for Cascaded Neural Networks

This paper aims to classify and locate objects accurately and efficiently, without using bounding box annotations. It is challenging as objects in the wild could appear at arbitrary locations and in different scales. In this paper, we propose a novel classification architecture ProNet based on convolutional neural networks. It uses computationally efficient neural networks to propose image regions that are likely to contain objects, and applies more powerful but slower networks on the proposed regions. The basic building block is a multi-scale fully-convolutional network which assigns object confidence scores to boxes at different locations and scales. We show that such networks can be trained effectively using image-level annotations, and can be connected into cascades or trees for efficient object classification. ProNet outperforms previous state-of-the-art significantly on PASCAL VOC 2012 and MS COCO datasets for object classification and point-based localization.

cs.CV↗
arXiv2015

Pati-Salam Unification from Non-commutative Geometry and the TeV-scale $W_R$ boson

We analyze the compatibility of the unified left-right symmetric Pati-Salam models motivated by non-commutative geometry and the TeV scale right-handed W boson suggested by recent LHC data. We find that the unification/matching conditions place conflicting demands on the symmetry breaking scales and that generating the required W_R mass and coupling is non-trivial.

hep-ph↗
arXiv2015

Automatic Concept Discovery from Parallel Text and Visual Corpora

Humans connect language and vision to perceive the world. How to build a similar connection for computers? One possible way is via visual concepts, which are text terms that relate to visually discriminative entities. We propose an automatic visual concept discovery algorithm using parallel text and visual corpora; it filters text terms based on the visual discriminative power of the associated images, and groups them into concepts using visual and semantic similarities. We illustrate the applications of the discovered concepts using bidirectional image and sentence retrieval task and image tagging task, and show that the discovered concepts not only outperform several large sets of manually selected concepts significantly, but also achieves the state-of-the-art performance in the retrieval task.

cs.CV↗
arXiv2015

Exact transition probabilities for a linear sweep through a Kramers-Kronig resonance

We consider a localized electronic spin controlled by a circularly polarized optical beam and an external magnetic field. When the frequency of the beam is tuned near an optical resonance with a continuum of higher energy states, effective magnetic fields are induced on the two-level system via the Inverse Faraday Effect. We explore the process in which the frequency of the beam is made linearly time-dependent so that it sweeps through the optical resonance, starting and ending at the values far away from it. In addition to changes of spin states, Kramers-Kronig relations guarantee that a localized electron can also escape into a continuum of states. We argue that probabilities of transitions between different possible electronic states after such a sweep of the optical frequency can be found exactly regardless the shape of the resonance. We also discuss extension of our results to multistate systems.

cond-mat.mes-hall↗
arXiv2015

Temporal Localization of Fine-Grained Actions in Videos by Domain Transfer from Web Images

We address the problem of fine-grained action localization from temporally untrimmed web videos. We assume that only weak video-level annotations are available for training. The goal is to use these weak labels to identify temporal segments corresponding to the actions, and learn models that generalize to unconstrained web videos. We find that web images queried by action names serve as well-localized highlights for many actions, but are noisily labeled. To solve this problem, we propose a simple yet effective method that takes weak video labels and noisy image labels as input, and generates localized action frames as output. This is achieved by cross-domain transfer between video frames and web images, using pre-trained deep convolutional neural networks. We then use the localized action frames to train action recognition models with long short-term memory networks. We collect a fine-grained sports action data set FGA-240 of more than 130,000 YouTube videos. It has 240 fine-grained actions under 85 sports activities. Convincing results are shown on the FGA-240 data set, as well as the THUMOS 2014 localization data set with untrimmed training videos.

cs.CV↗
arXiv2015

The Higgs Mass, Superconnections and the TeV-scale Left-Right Symmetric Model

We discuss the physical implications of formulating the Standard Model (SM) in terms of the superconnection formalism involving the superalgebra su(2/1). In particular, we discuss the prediction of the Higgs mass according to the formalism and point out that it is ~170 GeV, in clear disagreement with experiment. To remedy this problem, we extend the formalism to the superalgebra su(2/2), which extends the SM to the left-right symmetric model (LRSM) and accommodates a ~126 GeV Higgs. Both the SM in the su(2/1) case and the LRSM in the su(2/2) case are argued to emerge at ~4 TeV from an underlying theory in which the spacetime geometry is modified by the addition of a discrete extra dimension. The formulation of the exterior derivative in this model space suggests a deep connection between the modified geometry, which can be described in the language of non-commutative geometry (NCG), and the spontaneous breaking of the gauge symmetries. The implication is that spontaneous symmetry breaking could actually be geometric/quantum gravitational in nature. The non-decoupling phenomenon seen in the Higgs sector can then be reinterpreted in a new light as due to the mixing of low energy (SM) physics and high energy physics associated with quantum gravity, such as string theory. The phenomenology of a TeV scale LRSM is also discussed, and we argue that some exciting discoveries may await us at the LHC, and other near-future experiments.

hep-ph↗
arXiv2014

Spin Correlations in Quantum Wires

We consider theoretically spin correlations in an 1D quantum wire with Rashba-Dresselhaus spin-orbit interaction (RDI). The correlations of non-interacting electrons display electron-spin resonance at a frequency proportional to the RDI coupling. Interacting electrons on varying the direction of external magnetic field transit from the state of Luttinger liquid (LL) to the spin density wave (SDW) state. We show that the two-time total spin correlations of these states are significantly different. In the LL the projection of total spin to the direction of the RDI induced field is conserved and the corresponding correlator is equal to zero. The correlators of two components perpendicular to the RDI field display a sharp ESR driven by RDI induced intrinsic field. In contrast, in the SDW state the longitudinal projection of spin dominates, whereas the transverse components are suppressed. This prediction indicates a simple way for experimental diagnostic of the SDW in a quantum wire.

cond-mat.str-el↗
arXiv2013

Knotted and linked products of recombination on $T(2,n)\#T(2,m)$ substrates

We develop a topological model of site-specific recombination that applies to substrates which are the connected sum of two torus links of the form $T(2,n)\#T(2,m)$. Then we use our model to prove that all knots and links that can be produced by site-specific recombination on such substrates are contained in one of two families, which we illustrate.

math.GT↗
arXiv2013

Magnetic properties of a long, thin-walled ferromagnetic nanotube

We consider magnetic properties of a long, thin-walled ferromagnetic nanotube. We assume that the tube consists of isotropic homogeneous magnet whose spins interact via the exchange energy, the dipole-dipole interaction energy, and also interact with an external field via Zeeman energy. Possible stable states are the parallel state with the magnetization along the axis of the tube, and the vortex state with the magnetization along azimuthal direction. For a given material, which of them has lower energy depends on the value γ=R^2d/(L λ_x^2), where R is the radius of the tube, d is its thickness, L is its length and λ_x is an intrinsic scale of length characterizing the ration of exchange and dipolar interaction. At γ<1 the parallel state wins, otherwise the vortex state is stable. A domain wall in the middle of the tube is always energy unfavorable, but it can exist as a metastable structure. Near the ends of a tube magnetized parallel to the axis a half-domain structure transforming gradually the parallel magnetization to a vortex just at the edge of the tube is energy favorable. We also consider the equilibrium magnetization textures in an external magnetic field either parallel or perpendicular to the tube. Finally, magnetic fields produced by a nanotube and an array of tubes is analyzed.

cond-mat.mtrl-sci↗