arXiv subjects

Yu-Shen Liu

Publications and source records attributed to Yu-Shen Liu.

102 records · Page 6Linked to original sources

arXiv2019

3DViewGraph: Learning Global Features for 3D Shapes from A Graph of Unordered Views with Attention

Learning global features by aggregating information over multiple views has been shown to be effective for 3D shape analysis. For view aggregation in deep learning models, pooling has been applied extensively. However, pooling leads to a loss of the content within views, and the spatial relationship among views, which limits the discriminability of learned features. We propose 3DViewGraph to resolve this issue, which learns 3D global features by more effectively aggregating unordered views with attention. Specifically, unordered views taken around a shape are regarded as view nodes on a view graph. 3DViewGraph first learns a novel latent semantic mapping to project low-level view features into meaningful latent semantic embeddings in a lower dimensional space, which is spanned by latent semantic patterns. Then, the content and spatial information of each pair of view nodes are encoded by a novel spatial pattern correlation, where the correlation is computed among latent semantic patterns. Finally, all spatial pattern correlations are integrated with attention weights learned by a novel attention mechanism. This further increases the discriminability of learned features by highlighting the unordered view nodes with distinctive characteristics and depressing the ones with appearance ambiguity. We show that 3DViewGraph outperforms state-of-the-art methods under three large-scale benchmarks.

cs.CV↗

arXiv2019

Adversarial Cross-Modal Retrieval via Learning and Transferring Single-Modal Similarities

Cross-modal retrieval aims to retrieve relevant data across different modalities (e.g., texts vs. images). The common strategy is to apply element-wise constraints between manually labeled pair-wise items to guide the generators to learn the semantic relationships between the modalities, so that the similar items can be projected close to each other in the common representation subspace. However, such constraints often fail to preserve the semantic structure between unpaired but semantically similar items (e.g. the unpaired items with the same class label are more similar than items with different labels). To address the above problem, we propose a novel cross-modal similarity transferring (CMST) method to learn and preserve the semantic relationships between unpaired items in an unsupervised way. The key idea is to learn the quantitative similarities in single-modal representation subspace, and then transfer them to the common representation subspace to establish the semantic relationships between unpaired items across modalities. Experiments show that our method outperforms the state-of-the-art approaches both in the class-based and pair-based retrieval tasks.

cs.MM↗

arXiv2018

Point2Sequence: Learning the Shape Representation of 3D Point Clouds with an Attention-based Sequence to Sequence Network

Exploring contextual information in the local region is important for shape understanding and analysis. Existing studies often employ hand-crafted or explicit ways to encode contextual information of local regions. However, it is hard to capture fine-grained contextual information in hand-crafted or explicit manners, such as the correlation between different areas in a local region, which limits the discriminative ability of learned features. To resolve this issue, we propose a novel deep learning model for 3D point clouds, named Point2Sequence, to learn 3D shape features by capturing fine-grained contextual information in a novel implicit way. Point2Sequence employs a novel sequence learning model for point clouds to capture the correlations by aggregating multi-scale areas of each local region with attention. Specifically, Point2Sequence first learns the feature of each area scale in a local region. Then, it captures the correlation between area scales in the process of aggregating all area scales using a recurrent neural network (RNN) based encoder-decoder structure, where an attention mechanism is proposed to highlight the importance of different area scales. Experimental results show that Point2Sequence achieves state-of-the-art performance in shape classification and segmentation tasks.

cs.CV↗

arXiv2018

View Inter-Prediction GAN: Unsupervised Representation Learning for 3D Shapes by Learning Global Shape Memories to Support Local View Predictions

In this paper we present a novel unsupervised representation learning approach for 3D shapes, which is an important research challenge as it avoids the manual effort required for collecting supervised data. Our method trains an RNN-based neural network architecture to solve multiple view inter-prediction tasks for each shape. Given several nearby views of a shape, we define view inter-prediction as the task of predicting the center view between the input views, and reconstructing the input views in a low-level feature space. The key idea of our approach is to implement the shape representation as a shape-specific global memory that is shared between all local view inter-predictions for each shape. Intuitively, this memory enables the system to aggregate information that is useful to better solve the view inter-prediction tasks for each shape, and to leverage the memory as a view-independent shape representation. Our approach obtains the best results using a combination of L_2 and adversarial losses for the view inter-prediction task. We show that VIP-GAN outperforms state-of-the-art methods in unsupervised 3D feature learning on three large scale 3D shape benchmarks.

cs.CV↗

arXiv2018

Y^2Seq2Seq: Cross-Modal Representation Learning for 3D Shape and Text by Joint Reconstruction and Prediction of View and Word Sequences

A recent method employs 3D voxels to represent 3D shapes, but this limits the approach to low resolutions due to the computational cost caused by the cubic complexity of 3D voxels. Hence the method suffers from a lack of detailed geometry. To resolve this issue, we propose Y^2Seq2Seq, a view-based model, to learn cross-modal representations by joint reconstruction and prediction of view and word sequences. Specifically, the network architecture of Y^2Seq2Seq bridges the semantic meaning embedded in the two modalities by two coupled `Y' like sequence-to-sequence (Seq2Seq) structures. In addition, our novel hierarchical constraints further increase the discriminability of the cross-modal representations by employing more detailed discriminative information. Experimental results on cross-modal retrieval and 3D shape captioning show that Y^2Seq2Seq outperforms the state-of-the-art methods.

cs.CV↗

arXiv2011

Atomic-scale Field-effect Transistor as a Thermoelectric Power Generator and Self-powered Device

Using first-principles approaches, we have investigated the thermoelectric properties and the energy conversion efficiency of the paired metal-Br-Al junction. Owing to the narrow states in the vicinity of the chemical potential, the nanojunction has large Seebeck coefficients such that it can be considered an efficient thermoelectric power generator. We also consider the nanojunction in a three-terminal geometry, where the current, voltage, power, and efficiency can be efficiently modulated by the gate voltages. Such current-voltage characteristics could be useful in the design of nano-scale electronic devices such, as a transistor or switch. Notably, the nanojunction as a transistor with a fixed finite temperature difference between electrodes can power itself using the Seebeck effect.

cond-mat.mes-hall↗

arXiv2011

Counting Statistics in Nanoscale Junctions

We present first-principles calculations for moments of the current up to the third order in atomic-scale junctions. The quantum correlations of the current are calculated using the current operator in terms of the wave functions obtained self-consistently within the static density-functional theory. We investigate the relationships of the conductance, the second, and the third moment of the current for carbon atom chains of various lengths bridging two metal electrodes in the linear and nonlinear regimes. The conductance, the second-, and the third-order Fano factors exhibit odd-even oscillation with the number of carbon atoms due to the full and half filled π* orbital near the Fermi levels. The third-order Fano factor and the conductance are positively correlated.

cond-mat.mes-hall↗

arXiv2010

Effect of Thermoelectric Cooling in Nanoscale Junctions

We propose a thermoelectric cooling device based on an atomic-sized junction. Using first-principles approaches, we investigate the working conditions and the coefficient of performance (COP) of an atomic-scale electronic refrigerator where the effects of phonon's thermal current and local heating are included. It is observed that the functioning of the thermoelectric nano-refrigerator is restricted to a narrow range of driving voltages. Compared with the bulk thermoelectric system with the overwhelmingly irreversible Joule heating, the 4-Al atomic refrigerator has a higher efficiency than a bulk thermoelectric refrigerator with the same $ZT$ due to suppressed local heating via the quasi-ballistic electron transport and small driving voltages. Quantum nature due to the size minimization offered by atomic-level control of properties facilitates electron cooling beyond the expectation of the conventional thermoelectric device theory.

cond-mat.mes-hall↗

arXiv2010

Seebeck Coefficients in Nanoscale Junctions: Effects of Electron-vibration Scattering and Local Heating

We report first-principles calculations of inelastic Seebeck coefficients in an aluminum monatomic junction. We compare the elastic and inelastic Seebeck coefficients with and without local heating. In the low temperature regime, the signature of normal modes in the profiles of the inelastic Seebeck effects is salient. The inelastic Seebeck effects are enhanced by the normal modes, and further magnified by local heating. In the high temperature regime, the inelastic Seebeck effects are weakly suppressed due to the quasi-ballistic transport.

cond-mat.mes-hall↗

arXiv2009

Seebeck coefficient of thermoelectric moleculat junction: First-principles calculations

A first-principles approach is presented for the thermoelectricity in molecular junctions formed by a single molecule contact. The study investigates the Seebeck coefficient considering the source-drain electrodes with distinct temperatures and chemical potentials in a three-terminal geometry junction. We compare the Seebeck coefficient in the amino-substituted and unsubstituted butanethiol junction and observe interesting thermoelectric properties in the amino-substituted junction. Due to the novel states around the Fermi levels introduced by the amino-substitution, the Seebeck coefficient could be easily modulated by using gate voltages and biases. When the temperature in one of the electrodes is fixed, the Seebeck coefficient varies significantly with the temperature in the other electrode, and such dependence could be modulated by varying the gate voltages. As the biases increase, richer features in the Seebeck coefficient are observed, which are closely related to the transmission functions in the vicinity of the left and right Fermi levels.

cond-mat.mes-hall↗

arXiv2009

Do Thermoelectric Materials in Nanojunctions Display Material Property or Junction Property?

The miniaturization of thermoelectric nanojunctions raises a fundamental question: do the thermoelectric quantities of the bridging materials in nanojunctions remain to display material properties or show junction properties? In order to answer this question, we investigate the Seebeck coefficient $S$ and the thermoelectric figure of merit $ZT$ especially in relation to the length characteristics of the junctions from the first-principles approaches. For $S$, the metallic atomic chains reveal strong length characteristics related to strong hybridization in the electronic structures between the atoms and electrodes, while the insulating molecular wires display strong material properties due to the cancelation of exponential scalings in the DOSs. For $ZT$, the atomic wires remain to show strong junction properties. However, the length chrematistics of the insulation molecular wires depend on a characteristic temperature $T_{0}= \sqrt{β/γ(l)}$ around 10K. When $T \ll T_{0}$, where the electron transport dominates the thermal current, the molecular junctions remain to show material properties. When $T \gg T_{0}$, where the phonon transport dominates the thermal current, the molecular junctions display junction properties.

cond-mat.mes-hall↗

arXiv2009

Efficiency of Energy Conversion in Thermoelectric Nanojunctions

Using first-principles approaches, this study investigated the efficiency of energy conversion in nanojunctions, described by the thermoelectric figure of merit $ZT$. We obtained the qualitative and quantitative descriptions for the dependence of $ZT$ on temperatures and lengths. A characteristic temperature: $T_{0}= \sqrt{β/γ(l)}$ was observed. When $T\ll T_{0}$, $ZT\propto T^{2}$. When $T\gg T_{0}$, $ZT$ tends to a saturation value. The dependence of $ZT$ on the wire length for the metallic atomic chains is opposite to that for the insulating molecules: for aluminum atomic (conducting) wires, the saturation value of $ZT$ increases as the length increases; while for alkanethiol (insulating) chains, the saturation value of $ZT$ decreases as the length increases. $ZT$ can also be enhanced by choosing low-elasticity bridging materials or creating poor thermal contacts in nanojunctions. The results of this study may be of interest to research attempting to increase the efficiency of energy conversion in nano thermoelectric devices.

cond-mat.mes-hall↗