In this paper, we contribute a human evaluation metric for evaluating different styles on a generative model. We also evaluate eight different automated methods, finding that using Inception embeddings preceding the auxiliary classifier correlates m…

A methodology of building lane change model was proposed, the lane-changing decision lies on the drivers’ own incentives. Within multiple-criteria, the combination of incentives can easily realized and convenient to mimic different chauffeur type by…

Existing techniques for agreement analysis strongly rely on the formation of equivalence classes and ignore the integral properties that are shared between the gestures. In this work, we represent gestures as binary description vectors by decomposin…

In this work, we have introduced two models for inferring decompositional semantic graph structures from text: a pipeline system and a joint parsing-as-transduction model. The joint parser that we have introduced is comparable to or better than the …

In this paper, we have introduced a novel method to learn a global graph that explains the structure of a set of smooth signals using partial information provided by multi-layer graphs. In comparison to the state-of-the-art graph learning methods, t…

In this work, we propose a novel dual part-aligned representation scheme to address the non-human part misalignment problem for person re-identification. It consists of a human part branch and a latent part branch to tackle both human part misalignm…

By emphasizing on the role of non-stationarity in an interacting network, we introduced a novel approach for analyzing it. The main focus of our analysis was also on the problem of error diagnosis and canceling certain interferences caused by damage…

This paper presented an end-to-end approach for music genre classification based on 2D CNNs and evaluated it with two 2D representations: MF spectrograms and STFT spectrograms. The proposed approach learns from spectrograms of audio segments and per…

In this work, we introduced a signal processing-informed deep learning architecture for modeling artificial reverberators. We explored the capabilities of learning sparse FIR filters and time-varying mixing gains within a DNN framework. We show the …

In this work we start by investigating the practicability of the state-of-the-art stochastic gradient methods in imaging inverse problems where we chose image deblurring as a running example. We firstly present a surprisingly negative result on exis…

The 11,826 DNA oPools oligos were ordered from IDT (https://www.idtdna.com/pages/products/custom-dna-rna/dna-oligos/custom-dna-oligos/opools-oligo-pools). They were PCR amplified and the PCR products were then converted into a shotgun sequencing lib…

This work has extended the Bayesian online change-point detection method to more complex scenarios by considering a hierarchical model, which is based on latent-class variables. To prevent the limitation of fixing the order of the hierarchical model…

In this paper, we have studied matrix factorization methods for the rating prediction task. We have first argued that each user has her own views w.r.t. items and that a single common method/model is unlikely to satisfy all users. We have proposed a…

We showed promising results in estimating well-being status with voice in terms of various measurements, i.e. anxiety, sleep quality, and mood. The proposed method utilizes acoustic and linguistic features along with response- and measurement- level…

This study explores the use of a novel regularization method, minimum hyperspherical energy (MHE), for improving the task of singing voice separation in the Wave-U-Net. It is, to the authors’ knowledge, the first time that this technique is applied …

In this paper, we propose a model for action recognition that takes only RGB frames as input, and extracts motion features leveraging predicting coding networks (PredNets). We explored the combination of a spatial model ResNet-50 with PredNet for th…

We have shown that patch attacks generalize to state-of-the-art optical flow networks and can considerably impact the performance of optical flow systems that use deep neural networks. The patch attacks are invariant to translation and small changes…

Uncertainty quantification of complex and strongly corrupted data can be challenging, and Bayesian posterior analysis is an approach that can quantify complex posterior distributions such as multimodal solutions. It requires a reliable prior distrib…

When ML models do not meet any of the criteria imposed to declare them transparent, a separate method must be devised and applied to the model to explain its decisions. This is the purpose of post-hoc explainability techniques (also referred to as p…