We aim to automatically identify human action reasons in online videos. We focus on the widespread genre of lifestyle vlogs, in which people perform actions while verbally describing them. We introduce and make publicly available the WhyAct dataset, consisting of 1,077 visual actions manually annotated with their reasons. We describe a multimodal model that leverages visual and textual information to automatically infer the reasons corresponding to an action presented in the video.
This paper investigates a new model to improve the scalability of low-power long-range (LoRa) networks by allowing multiple end devices (EDs) to simultaneously communicate with multiple multi-antenna gateways on the same frequency band and using the same spreading factor. The maximum likelihood (ML) decision rule is first derived for non-coherent detection of information bits transmitted by multiple devices. To overcome the high complexity of the ML detection, we propose a sub-optimal two-stage detection algorithm to balance the computational complexity and error performance. In the first stage, we identify transmit chirps (without knowing which EDs transmit them). In the second stage, we determine the EDs that transmit the specific chirps identified from the first stage. To improve the detection performance in the second stage, we also optimize the transmit powers of EDs to minimize the similarity, measured by the Jaccard coefficient, between the received powers of any pair of EDs. As the power control optimization problem is non-convex, we use concepts from successive convex approximation to transform it to an approximate convex optimization problem that can be solved iteratively and guaranteed to reach a sub-optimal solution. Simulation results demonstrate and justify the tradeoff between transmit power penalties and network scalability of the proposed LoRa network model. In particular, by allowing concurrent transmission of 2 or 3 EDs, the uplink capacity of the proposed network can be doubled or tripled over that of a conventional LoRa network, albeit at the expense of additional 3.0 or 4.7 dB transmit power.
With the advancement of computer technology, there is a surge of interest in effective mapping methods for objects in higher-dimensional spaces. To establish a one-to-one correspondence between objects, higher-dimensional quasi-conformal theory can be utilized for ensuring the bijectivity of the mappings. In addition, it is often desirable for the mappings to satisfy certain prescribed geometric constraints and possess low distortion in conformality or volume. In this work, we develop a unifying framework for computing $n$-dimensional quasi-conformal mappings. More specifically, we propose a variational model that integrates quasi-conformal distortion, volumetric distortion, landmark correspondence, intensity mismatch and volume prior information to handle a large variety of deformation problems. We further prove the existence of a minimizer for the proposed model and devise efficient numerical methods to solve the optimization problem. We demonstrate the effectiveness of the proposed framework using various experiments in two- and three-dimensions, with applications to medical image registration, adaptive remeshing and shape modeling.
Partial Multi-label Learning (PML) aims to induce the multi-label predictor from datasets with noisy supervision, where each training instance is associated with several candidate labels but only partially valid. To address the noisy issue, the existing PML methods basically recover the ground-truth labels by leveraging the ground-truth confidence of the candidate label, \ie the likelihood of a candidate label being a ground-truth one. However, they neglect the information from non-candidate labels, which potentially contributes to the ground-truth label recovery. In this paper, we propose to recover the ground-truth labels, \ie estimating the ground-truth confidences, from the label enrichment, composed of the relevance degrees of candidate labels and irrelevance degrees of non-candidate labels. Upon this observation, we further develop a novel two-stage PML method, namely \emph{\underline{P}artial \underline{M}ulti-\underline{L}abel \underline{L}earning with \underline{L}abel \underline{E}nrichment-\underline{R}ecovery} (\baby), where in the first stage, it estimates the label enrichment with unconstrained label propagation, then jointly learns the ground-truth confidence and multi-label predictor given the label enrichment. Experimental results validate that \baby outperforms the state-of-the-art PML methods.
Contextualised word embeddings generated from Neural Language Models (NLMs), such as BERT, represent a word with a vector that considers the semantics of the target word as well its context. On the other hand, static word embeddings such as GloVe represent words by relatively low-dimensional, memory- and compute-efficient vectors but are not sensitive to the different senses of the word. We propose Context Derived Embeddings of Senses (CDES), a method that extracts sense related information from contextualised embeddings and injects it into static embeddings to create sense-specific static embeddings. Experimental results on multiple benchmarks for word sense disambiguation and sense discrimination tasks show that CDES can accurately learn sense-specific static embeddings reporting comparable performance to the current state-of-the-art sense embeddings.
Automatic speaker verification (ASV) systems, which determine whether two speeches are from the same speaker, mainly focus on verification accuracy while ignoring inference speed. However, in real applications, both inference speed and verification accuracy are essential. This study proposes cross-sequential re-parameterization (CS-Rep), a novel topology re-parameterization strategy for multi-type networks, to increase the inference speed and verification accuracy of models. CS-Rep solves the problem that existing re-parameterization methods are unsuitable for typical ASV backbones. When a model applies CS-Rep, the training-period network utilizes a multi-branch topology to capture speaker information, whereas the inference-period model converts to a time-delay neural network (TDNN)-like plain backbone with stacked TDNN layers to achieve the fast inference speed. Based on CS-Rep, an improved TDNN with friendly test and deployment called Rep-TDNN is proposed. Compared with the state-of-the-art model ECAPA-TDNN, which is highly recognized in the industry, Rep-TDNN increases the actual inference speed by about 50% and reduces the EER by 10%. The code will be released.
Advancements in attention mechanisms have led to significant performance improvements in a variety of areas in machine learning due to its ability to enable the dynamic modeling of temporal sequences. A particular area in computer vision that is likely to benefit greatly from the incorporation of attention mechanisms in video action recognition. However, much of the current research's focus on attention mechanisms have been on spatial and temporal attention, which are unable to take advantage of the inherent motion found in videos. Motivated by this, we develop a new attention mechanism called Motion Aware Attention (M2A) that explicitly incorporates motion characteristics. More specifically, M2A extracts motion information between consecutive frames and utilizes attention to focus on the motion patterns found across frames to accurately recognize actions in videos. The proposed M2A mechanism is simple to implement and can be easily incorporated into any neural network backbone architecture. We show that incorporating motion mechanisms with attention mechanisms using the proposed M2A mechanism can lead to a +15% to +26% improvement in top-1 accuracy across different backbone architectures, with only a small increase in computational complexity. We further compared the performance of M2A with other state-of-the-art motion and attention mechanisms on the Something-Something V1 video action recognition benchmark. Experimental results showed that M2A can lead to further improvements when combined with other temporal mechanisms and that it outperforms other motion-only or attention-only mechanisms by as much as +60% in top-1 accuracy for specific classes in the benchmark.
Fully automatic opera tracking is challenging because of the acoustic complexity of the genre, combining musical and linguistic information (singing, speech) in complex ways. In this paper, we propose a new pipeline for complete opera tracking. The pipeline is based on two trackers. A music tracker that has proven to be effective at tracking orchestral parts, will lead the tracking process. In addition, a lyrics tracker, that has recently been shown to reliably track the lyrics of opera songs, will correct the music tracker when tracking parts that have a text dominance over the music. We will demonstrate the efficiency of this method on the opera Don Giovanni, showing that this technique helps improving accuracy and robustness of a complete opera tracker.
We propose a method for learning de-identified prosody representations from raw audio using a contrastive self-supervised signal. Whereas prior work has relied on conditioning models on bottlenecks, we introduce a set of inductive biases that exploit the natural structure of prosody to minimize timbral information and decouple prosody from speaker representations. Despite aggressive downsampling of the input and having no access to linguistic information, our model performs comparably to state-of-the-art speech representations on DAMMP, a new benchmark we introduce for spoken language understanding. We use minimum description length probing to show that our representations have selectively learned the subcomponents of non-timbral prosody, and that the product quantizer naturally disentangles them without using bottlenecks. We derive an information-theoretic definition of speech de-identifiability and use it to demonstrate that our prosody representations are less identifiable than other speech representations.
Shapelet-based algorithms are widely used for time series classification because of their ease of interpretation, but they are currently outperformed, notably by methods using convolutional kernels, capable of reaching state-of-the-art performance while being highly scalable. We present a new formulation of time series shapelets including the notion of dilation, and a shapelet extraction method based on convolutional kernels, which is able to target the discriminant information identified by convolutional kernels. Experiments performed on 108 datasets show that our method improves on the state-of-the-art for shapelet algorithms, and we show that it can be used to interpret results from convolutional kernels.