hSMAL: Detailed Horse Shape and Pose Reconstruction for Motion Pattern Recognition

In this paper we present our preliminary work on model-based behavioral analysis of horse motion. Our approach is based on the SMAL model, a 3D articulated statistical model of animal shape. We define a novel SMAL model for horses based on a new template, skeleton and shape space learned from $37$ horse toys. We test the accuracy of our hSMAL model in reconstructing a horse from 3D mocap data and images. We apply the hSMAL model to the problem of lameness detection from video, where we fit the model to images to recover 3D pose and train an ST-GCN network on pose data. A comparison with the same network trained on mocap points illustrates the benefit of our approach.

Imaging the breaking of electrostatic dams in graphene for ballistic and viscous fluids

The flow of charge carriers in materials can, under some circumstances, mimic the flow of viscous fluids. In order to visualize the consequences of such effects, new methodologies must be developed that can probe the quasiparticle flow profile with nm-scale resolution as the geometric parameters of the system are continuously evolved. In this work, scanning tunneling potentiometry (STP) is used to image quasiparticle flow around engineered electrostatic barriers in graphene/hBN heterostructures. Measurements are performed as electrostatic dams - defined by lateral pn-junction barriers - are broken within the graphene sheet, and carriers move through conduction channels with physical widths that vary continuously from pinch-off to um-scale. Local, STP measurements of the electrochemical potential allow for direct characterization of the evolving flow profile, which we compare to finite-element simulations of a Stokesian fluid with varying parameters. Our results reveal distinctly non-Ohmic flow profiles, with charge dipoles forming across barriers due to carrier scattering and accumulation on the upstream side, and depletion downstream. Conductance measurements of individual channels, meanwhile, reveal that at low temperatures the quasiparticle flow is ballistic, but as the temperature is raised there is a Knudsen-to-Gurzhi regime crossover where the fluid becomes viscous and the channel conductance exceeds the ballistic limit set by Sharvin conductance. These results provide a clear illustration of how carrier flow in a Fermi fluid evolves as a function of carrier density, channel width, and temperature. They also demonstrate how STP can be used to extract key parameters of quasiparticle transport, with a spatial resolution that exceeds that of other methods by orders of magnitude.

Conformal maps of viscous electron flow in the Gurzhi crossover

We investigate the impact of geometric constriction on the viscous flow of electron liquid through quantum point contacts. We provide analysis on the electric potential distribution given the setup of a slit configuration and use the method of conformal mapping to obtain analytical results. The potential profile can be tested and contrasted experimentally with the scanning tunneling potentiometry technique. We discuss intricate physics that underlies the Gurzhi effect, i.e. the enhancement of conductivity in the viscous flow, and calculate the temperature dependence of the momentum relaxation time as a result of impurity assisted quasi-ballistic interference effects. We caution that spatially inhomogeneous profiles of current in the Gurzhi crossover between Ohmic and Stokes flows might also appear in the non-hydrodynamic regime where non-locality plays an important role.

Revealing the non-adiabatic and non-Abelian multiple-band effects via anisotropic valley Hall conduction in bilayer graphene

Many quantum materials of interest, ex., bilayer graphene, possess a number of closely spaced but not fully degenerate bands near the Fermi level, where the coupling to the far detuned remote bands can induce Berry curvatures of the non-Abelian character in this active multiple-band manifold for transport effects. Under finite electric fields, non-adiabatic interband transition processes are expected to play significant roles in the associated Hall conduction. Here through an exemplified study on the valley Hall conduction in AB-stacked bilayer graphene, we show that the contribution arising from non-adiabatic transitions around the bands near the Fermi energy to the Hall current is not only quantitatively about an order-of-magnitude larger than the contribution due to adiabatic inter-manifold transition with the non-Abelian Berry curvatures. Due to the trigonal warping, the former also displays an anisotropic response to the orientation of the applied electric field that is qualitatively distinct from that of the latter. We further show that these anisotropic responses also reveal the essential differences between the diagonal and off-diagonal elements of the non-Abelian Berry curvature matrix in terms of their contributions to the Hall currents. We provide a physically intuitive understanding of the origin of distinct anisotropic features from different Hall current contributions, in terms of band occupations and interband coherence. This then points to the generalization beyond the specific example of bilayer graphenes.

Transfer Learning for Piano Sustain-Pedal Detection

Detecting piano pedalling techniques in polyphonic music remains a challenging task in music information retrieval. While other piano-related tasks, such as pitch estimation and onset detection, have seen improvement through applying deep learning methods, little work has been done to develop deep learning models to detect playing techniques. In this paper, we propose a transfer learning approach for the detection of sustain-pedal techniques, which are commonly used by pianists to enrich the sound. In the source task, a convolutional neural network (CNN) is trained for learning spectral and temporal contexts when the sustain pedal is pressed using a large dataset generated by a physical modelling virtual instrument. The CNN is designed and experimented through exploiting the knowledge of piano acoustics and physics. This can achieve an accuracy score of 0.98 in the validation results. In the target task, the knowledge learned from the synthesised data can be transferred to detect the sustain pedal in acoustic piano recordings. A concatenated feature vector using the activations of the trained convolutional layers is extracted from the recordings and classified into frame-wise pedal press or release. We demonstrate the effectiveness of our method in acoustic piano recordings of Chopins music. From the cross-validation results, the proposed transfer learning method achieves an average F-measure of 0.89 and an overall performance of 0.84 obtained using the micro-averaged F-measure. These results outperform applying the pre-trained CNN model directly or the model with a fine-tuned last layer.

A Paragraph-level Multi-task Learning Model for Scientific Fact-Verification

Even for domain experts, it is a non-trivial task to verify a scientific claim by providing supporting or refuting evidence rationales. The situation worsens as misinformation is proliferated on social media or news websites, manually or programmatically, at every moment. As a result, an automatic fact-verification tool becomes crucial for combating the spread of misinformation. In this work, we propose a novel, paragraph-level, multi-task learning model for the SciFact task by directly computing a sequence of contextualized sentence embeddings from a BERT model and jointly training the model on rationale selection and stance prediction.

Context-aware Stand-alone Neural Spelling Correction

Existing natural language processing systems are vulnerable to noisy inputs resulting from misspellings. On the contrary, humans can easily infer the corresponding correct words from their misspellings and surrounding context. Inspired by this, we address the stand-alone spelling correction problem, which only corrects the spelling of each token without additional token insertion or deletion, by utilizing both spelling information and global context representations. We present a simple yet powerful solution that jointly detects and corrects misspellings as a sequence labeling task by fine-turning a pre-trained language model. Our solution outperforms the previous state-of-the-art result by 12.8% absolute F0.5 score.

Learning Audio Embeddings with User Listening Data for Content-based Music Recommendation

Personalized recommendation on new track releases has always been a challenging problem in the music industry. To combat this problem, we first explore user listening history and demographics to construct a user embedding representing the users music preference. With the user embedding and audio data from users liked and disliked tracks, an audio embedding can be obtained for each track using metric learning with Siamese networks. For a new track, we can decide the best group of users to recommend by computing the similarity between the tracks audio embedding and different user embeddings, respectively. The proposed system yields state-of-the-art performance on content-based music recommendation tested with millions of users and tracks. Also, we extract audio embeddings as features for music genre classification tasks. The results show the generalization ability of our audio embeddings.

Flows Succeed Where GANs Fail: Lessons from Low-Dimensional Data

Normalizing flows and generative adversarial networks (GANs) are both approaches to density estimation that use deep neural networks to transform samples from an uninformative prior distribution to an approximation of the data distribution. There is great interest in both for general-purpose statistical modeling, but the two approaches have seldom been compared to each other for modeling non-image data. The difficulty of computing likelihoods with GANs, which are implicit models, makes conducting such a comparison challenging. We work around this difficulty by considering several low-dimensional synthetic datasets. An extensive grid search over GAN architectures, hyperparameters, and training procedures suggests that no GAN is capable of modeling our simple low-dimensional data well, a task we view as a prerequisite for an approach to be considered suitable for general-purpose statistical modeling. Several normalizing flows, on the other hand, excelled at these tasks, even substantially outperforming WGAN in terms of Wasserstein distance---the metric that WGAN alone targets. Overall, normalizing flows appear to be more reliable tools for statistical inference than GANs.

Hydrodynamic electron transport near charge neutrality

We develop the theory of hydrodynamic electron transport in a long-range disorder potential for conductors in which the underlying electron liquid lacks Galilean invariance. For weak disorder, we express the transport coefficients of the system in terms of the intrinsic kinetic coefficients of the electron liquid and the correlation function of the disorder potential. We apply these results to analyze the doping and temperature dependence of transport coefficients of graphene devices. We show that at charge neutrality, long-range disorder increases the conductivity of the system above the intrinsic value. The enhancement arises from the predominantly vortical hydrodynamic flow caused by local deviations from charge neutrality. Its magnitude is inversely proportional to the shear viscosity of the electron liquid and scales as the square of the disorder correlation radius. This is qualitatively different from the situation away from charge neutrality. In that case, the flow is predominantly potential, and produces negative viscous contributions to the conductivity, which are proportional to the sum of shear and bulk viscosities, and inversely proportional to the square of disorder correlation radius.