Donghuo Zeng

Monisha Singh, Ximi Hoque, Donghuo Zeng et al.

The degree of concentration, enthusiasm, optimism, and passion displayed by individual(s) while interacting with a machine is referred to as `user engagement'. Engagement comprises of behavioral, cognitive, and affect related cues. To create engagement prediction systems that can work in real-world conditions, it is quintessential to learn from rich, diverse datasets. To this end, a large scale multi-faceted engagement in the wild dataset EngageNet is proposed. 31 hours duration data of 127 participants representing different illumination conditions are recorded. Thorough experiments are performed exploring the applicability of different features, action units, eye gaze, head pose, and MARLIN. Data from user interactions (question-answer) are analyzed to understand the relationship between effective learning and user engagement. To further validate the rich nature of the dataset, evaluation is also performed on the EngageWild dataset. The experiments show the usefulness of the proposed dataset. The code, models, and dataset link are publicly available at https://github.com/engagenet/engagenet_baselines.

Donghuo Zeng, Yanan Wang, Jianming Wu et al.

The heterogeneity gap problem is the main challenge in cross-modal retrieval. Because cross-modal data (e.g. audiovisual) have different distributions and representations that cannot be directly compared. To bridge the gap between audiovisual modalities, we learn a common subspace for them by utilizing the intrinsic correlation in the natural synchronization of audio-visual data with the aid of annotated labels. TNN-CCCA is the best audio-visual cross-modal retrieval (AV-CMR) model so far, but the model training is sensitive to hard negative samples when learning common subspace by applying triplet loss to predict the relative distance between inputs. In this paper, to reduce the interference of hard negative samples in representation learning, we propose a new AV-CMR model to optimize semantic features by directly predicting labels and then measuring the intrinsic correlation between audio-visual data using complete cross-triple loss. In particular, our model projects audio-visual features into label space by minimizing the distance between predicted label features after feature projection and ground label representations. Moreover, we adopt complete cross-triplet loss to optimize the predicted label features by leveraging the relationship between all possible similarity and dissimilarity semantic information across modalities. The extensive experimental results on two audio-visual double-checked datasets have shown an improvement of approximately 2.1% in terms of average MAP over the current state-of-the-art method TNN-CCCA for the AV-CMR task, which indicates the effectiveness of our proposed model.

Donghuo Zeng, Jianming Wu, Yanan Wang et al.

Large-scale open-domain dialogue systems such as PLATO-2 have achieved state-of-the-art scores in both English and Chinese. However, little work explores whether such dialogue systems also work well in the Japanese language. In this work, we create a large-scale Japanese dialogue dataset, Dialogue-Graph, which contains 1.656 million dialogue data in a tree structure from News, TV subtitles, and Wikipedia corpus. Then, we train PLATO-2 using Dialogue-Graph to build a large-scale Japanese dialogue system, PLATO-JDS. In addition, to improve the PLATO-JDS in the topic switch issue, we introduce a topic-switch algorithm composed of a topic discriminator to switch to a new topic when user input differs from the previous topic. We evaluate the user experience by using our model with respect to four metrics, namely, coherence, informativeness, engagingness, and humanness. As a result, our proposed PLATO-JDS achieves an average score of 1.500 for the human evaluation with human-bot chat strategy, which is close to the maximum score of 2.000 and suggests the high-quality dialogue generation capability of PLATO-2 in Japanese. Furthermore, our proposed topic-switch algorithm achieves an average score of 1.767 and outperforms PLATO-JDS by 0.267, indicating its effectiveness in improving the user experience of our system.

Yanan Wang, Donghuo Zeng, Shinya Wada et al.

Multimodal transfer learning aims to transform pretrained representations of diverse modalities into a common domain space for effective multimodal fusion. However, conventional systems are typically built on the assumption that all modalities exist, and the lack of modalities always leads to poor inference performance. Furthermore, extracting pretrained embeddings for all modalities is computationally inefficient for inference. In this work, to achieve high efficiency-performance multimodal transfer learning, we propose VideoAdviser, a video knowledge distillation method to transfer multimodal knowledge of video-enhanced prompts from a multimodal fundamental model (teacher) to a specific modal fundamental model (student). With an intuition that the best learning performance comes with professional advisers and smart students, we use a CLIP-based teacher model to provide expressive multimodal knowledge supervision signals to a RoBERTa-based student model via optimizing a step-distillation objective loss -- first step: the teacher distills multimodal knowledge of video-enhanced prompts from classification logits to a regression logit -- second step: the multimodal knowledge is distilled from the regression logit of the teacher to the student. We evaluate our method in two challenging multimodal tasks: video-level sentiment analysis (MOSI and MOSEI datasets) and audio-visual retrieval (VEGAS dataset). The student (requiring only the text modality as input) achieves an MAE score improvement of up to 12.3% for MOSI and MOSEI. Our method further enhances the state-of-the-art method by 3.4% mAP score for VEGAS without additional computations for inference. These results suggest the strengths of our method for achieving high efficiency-performance multimodal transfer learning.

Donghuo Zeng, Kazushi Ikeda

The cross-modal retrieval model leverages the potential of triple loss optimization to learn robust embedding spaces. However, existing methods often train these models in a singular pass, overlooking the distinction between semi-hard and hard triples in the optimization process. The oversight of not distinguishing between semi-hard and hard triples leads to suboptimal model performance. In this paper, we introduce a novel approach rooted in curriculum learning to address this problem. We propose a two-stage training paradigm that guides the model's learning process from semi-hard to hard triplets. In the first stage, the model is trained with a set of semi-hard triplets, starting from a low-loss base. Subsequently, in the second stage, we augment the embeddings using an interpolation technique. This process identifies potential hard negatives, alleviating issues arising from high-loss functions due to a scarcity of hard triples. Our approach then applies hard triplet mining in the augmented embedding space to further optimize the model. Extensive experimental results conducted on two audio-visual datasets show a significant improvement of approximately 9.8% in terms of average Mean Average Precision (MAP) over the current state-of-the-art method, MSNSCA, for the Audio-Visual Cross-Modal Retrieval (AV-CMR) task on the AVE dataset, indicating the effectiveness of our proposed method.

Donghuo Zeng, Yi Yu, Keizo Oyama

This work present a music dataset named MusicTM-Dataset, which is utilized in improving the representation learning ability of different types of cross-modal retrieval (CMR). Little large music dataset including three modalities is available for learning representations for CMR. To collect a music dataset, we expand the original musical notation to synthesize audio and generated sheet-music image, and build musical notation based sheet-music image, audio clip and syllable-denotation text as fine-grained alignment, such that the MusicTM-Dataset can be exploited to receive shared representation for multimodal data points. The MusicTM-Dataset presents 3 kinds of modalities, which consists of the image of sheet-music, the text of lyrics and synthesized audio, their representations are extracted by some advanced models. In this paper, we introduce the background of music dataset and express the process of our data collection. Based on our dataset, we achieve some basic methods for CMR tasks. The MusicTM-Dataset are accessible in https: //github.com/dddzeng/MusicTM-Dataset.

Yanan Wang, Shuichiro Haruta, Donghuo Zeng et al.

Video question answering (VideoQA) is a task to predict the correct answer to questions posed about a given video. The system must comprehend spatial and temporal relationships among objects extracted from videos to perform causal and temporal reasoning. While prior works have focused on modeling individual object movements using transformer-based methods, they falter when capturing complex scenarios involving multiple objects (e.g., "a boy is throwing a ball in a hoop"). We propose a contrastive language event graph representation learning method called CLanG to address this limitation. Aiming to capture event representations associated with multiple objects, our method employs a multi-layer GNN-cluster module for adversarial graph representation learning, enabling contrastive learning between the question text and its relevant multi-object event graph. Our method outperforms a strong baseline, achieving up to 2.2% higher accuracy on two challenging VideoQA datasets, NExT-QA and TGIF-QA-R. In particular, it is 2.8% better than baselines in handling causal and temporal questions, highlighting its strength in reasoning multiple object-based events.

Yanan Wang, Shuichiro Haruta, Donghuo Zeng et al.

Capturing complex hierarchical human activities, from atomic actions (e.g., picking up one present, moving to the sofa, unwrapping the present) to contextual events (e.g., celebrating Christmas) is crucial for achieving high-performance video question answering (VideoQA). Recent works have expanded multimodal models (e.g., CLIP, LLaVA) to process continuous video sequences, enhancing the model's temporal reasoning capabilities. However, these approaches often fail to capture contextual events that can be decomposed into multiple atomic actions non-continuously distributed over relatively long-term sequences. In this paper, to leverage the spatial visual context representation capability of the CLIP model for obtaining non-continuous visual representations in terms of contextual events in videos, we convert long-term video sequences into a spatial image domain and finetune the multimodal model LLaVA for the VideoQA task. Our approach achieves competitive performance on the STAR task, in particular, with a 78.4% accuracy score, exceeding the current state-of-the-art score by 2.8 points on the NExTQA task.

Donghuo Zeng, Hao Niu, Zhi Li et al.

Metric learning is central to retrieval, yet its effects on embedding geometry and optimization dynamics are not well understood. We introduce a diagnostic framework, VARIANCE (intra-/inter-class variance) and GREEDINESS (active ratio and gradient norms), to compare seven representative losses, i.e., Contrastive, Triplet, N-pair, InfoNCE, ArcFace, SCL, and CCL, across five image-retrieval datasets. Our analysis reveals that Triplet and SCL preserve higher within-class variance and clearer inter-class margins, leading to stronger top-1 retrieval in fine-grained settings. In contrast, Contrastive and InfoNCE compact embeddings are achieved quickly through many small updates, accelerating convergence but potentially oversimplifying class structures. N-pair achieves a large mean separation but with uneven spacing. These insights reveal a form of efficiency-granularity trade-off and provide practical guidance: prefer Triplet/SCL when diversity preservation and hard-sample discrimination are critical, and Contrastive/InfoNCE when faster embedding compaction is desired.

Donghuo Zeng, Roberto Legaspi, Yuewen Sun et al.

Tailoring persuasive conversations to users leads to more effective persuasion. However, existing dialogue systems often struggle to adapt to dynamically evolving user states. This paper presents a novel method that leverages causal discovery and counterfactual reasoning for optimizing system persuasion capability and outcomes. We employ the Greedy Relaxation of the Sparsest Permutation (GRaSP) algorithm to identify causal relationships between user and system utterance strategies, treating user strategies as states and system strategies as actions. GRaSP identifies user strategies as causal factors influencing system responses, which inform Bidirectional Conditional Generative Adversarial Networks (BiCoGAN) in generating counterfactual utterances for the system. Subsequently, we use the Dueling Double Deep Q-Network (D3QN) model to utilize counterfactual data to determine the best policy for selecting system utterances. Our experiments with the PersuasionForGood dataset show measurable improvements in persuasion outcomes using our approach over baseline methods. The observed increase in cumulative rewards and Q-values highlights the effectiveness of causal discovery in enhancing counterfactual reasoning and optimizing reinforcement learning policies for online dialogue systems.

Donghuo Zeng, Roberto S. Legaspi, Yuewen Sun et al.

Customizing persuasive conversations related to the outcome of interest for specific users achieves better persuasion results. However, existing persuasive conversation systems rely on persuasive strategies and encounter challenges in dynamically adjusting dialogues to suit the evolving states of individual users during interactions. This limitation restricts the system's ability to deliver flexible or dynamic conversations and achieve suboptimal persuasion outcomes. In this paper, we present a novel approach that tracks a user's latent personality dimensions (LPDs) during ongoing persuasion conversation and generates tailored counterfactual utterances based on these LPDs to optimize the overall persuasion outcome. In particular, our proposed method leverages a Bi-directional Generative Adversarial Network (BiCoGAN) in tandem with a Dialogue-based Personality Prediction Regression (DPPR) model to generate counterfactual data. This enables the system to formulate alternative persuasive utterances that are more suited to the user. Subsequently, we utilize the D3QN model to learn policies for optimized selection of system utterances on counterfactual data. Experimental results we obtained from using the PersuasionForGood dataset demonstrate the superiority of our approach over the existing method, BiCoGAN. The cumulative rewards and Q-values produced by our method surpass ground truth benchmarks, showcasing the efficacy of employing counterfactual reasoning and LPDs to optimize reinforcement learning policy in online interactions.

Donghuo Zeng, Yanan Wang, Kazushi Ikeda et al.

Metric learning minimizes the gap between similar (positive) pairs of data points and increases the separation of dissimilar (negative) pairs, aiming at capturing the underlying data structure and enhancing the performance of tasks like audio-visual cross-modal retrieval (AV-CMR). Recent works employ sampling methods to select impactful data points from the embedding space during training. However, the model training fails to fully explore the space due to the scarcity of training data points, resulting in an incomplete representation of the overall positive and negative distributions. In this paper, we propose an innovative Anchor-aware Deep Metric Learning (AADML) method to address this challenge by uncovering the underlying correlations among existing data points, which enhances the quality of the shared embedding space. Specifically, our method establishes a correlation graph-based manifold structure by considering the dependencies between each sample as the anchor and its semantically similar samples. Through dynamic weighting of the correlations within this underlying manifold structure using an attention-driven mechanism, Anchor Awareness (AA) scores are obtained for each anchor. These AA scores serve as data proxies to compute relative distances in metric learning approaches. Extensive experiments conducted on two audio-visual benchmark datasets demonstrate the effectiveness of our proposed AADML method, significantly surpassing state-of-the-art models. Furthermore, we investigate the integration of AA proxies with various metric learning methods, further highlighting the efficacy of our approach.

Donghuo Zeng, Hao Niu, Masato Taya

Learning aligned multimodal embeddings from weakly paired, label-free corpora is challenging: pipelines often provide only pre-extracted features, clips contain multiple events, and spurious co-occurrences. We propose HSC-MAE (Hierarchical Semantic Correlation-Aware Masked Autoencoder), a dual-path teacher-student framework that enforces semantic consistency across three complementary levels of representation - from coarse to fine: (i) global-level canonical-geometry correlation via DCCA, which aligns audio and visual embeddings within a shared modality-invariant subspace; (ii) local-level neighborhood-semantics correlation via teacher-mined soft top-k affinities, which preserves multi-positive relational structure among semantically similar instances; and (iii) sample-level conditional-sufficiency correlation via masked autoencoding, which ensures individual embeddings retain discriminative semantic content under partial observation. Concretely, a student MAE path is trained with masked feature reconstruction and affinity-weighted soft top-k InfoNCE; an EMA teacher operating on unmasked inputs via the CCA path supplies stable canonical geometry and soft positives. Learnable multi-task weights reconcile competing objectives, and an optional distillation loss transfers teacher geometry into the student. Experiments on AVE and VEGAS demonstrate substantial mAP improvements over strong unsupervised baselines, validating that HSC-MAE yields robust and well-structured audio-visual representations.

Donghuo Zeng

Contrastive loss and triplet loss are widely used objectives in deep metric learning, yet their effects on representation quality remain insufficiently understood. We present a theoretical and empirical comparison of these losses, focusing on intra- and inter-class variance and optimization behavior (e.g., greedy updates). Through task-specific experiments with consistent settings on synthetic data and real datasets-MNIST, CIFAR-10-it is shown that triplet loss preserves greater variance within and across classes, supporting finer-grained distinctions in the learned representations. In contrast, contrastive loss tends to compact intra-class embeddings, which may obscure subtle semantic differences. To better understand their optimization dynamics, By examining loss-decay rate, active ratio, and gradient norm, we find that contrastive loss drives many small updates early on, while triplet loss produces fewer but stronger updates that sustain learning on hard examples. Finally, across both classification and retrieval tasks on MNIST, CIFAR-10, CUB-200, and CARS196 datasets, our results consistently show that triplet loss yields superior performance, which suggests using triplet loss for detail retention and hard-sample focus, and contrastive loss for smoother, broad-based embedding refinement.

Donghuo Zeng, Roberto Legaspi, Yuewen Sun et al.

We hypothesize that optimal system responses emerge from adaptive strategies grounded in causal and counterfactual knowledge. Counterfactual inference allows us to create hypothetical scenarios to examine the effects of alternative system responses. We enhance this process through causal discovery, which identifies the strategies informed by the underlying causal structure that govern system behaviors. Moreover, we consider the psychological constructs and unobservable noises that might be influencing user-system interactions as latent factors. We show that these factors can be effectively estimated. We employ causal discovery to identify strategy-level causal relationships among user and system utterances, guiding the generation of personalized counterfactual dialogues. We model the user utterance strategies as causal factors, enabling system strategies to be treated as counterfactual actions. Furthermore, we optimize policies for selecting system responses based on counterfactual data. Our results using a real-world dataset on social good demonstrate significant improvements in persuasive system outcomes, with increased cumulative rewards validating the efficacy of causal discovery in guiding personalized counterfactual inference and optimizing dialogue policies for a persuasive dialogue system.

Donghuo Zeng, Kazushi Ikeda

Metric learning projects samples into an embedded space, where similarities and dissimilarities are quantified based on their learned representations. However, existing methods often rely on label-guided representation learning, where representations of different modalities, such as audio and visual data, are aligned based on annotated labels. This approach tends to underutilize latent complex features and potential relationships inherent in the distributions of audio and visual data that are not directly tied to the labels, resulting in suboptimal performance in audio-visual embedding learning. To address this issue, we propose a novel architecture that integrates cross-modal triplet loss with progressive self-distillation. Our method enhances representation learning by leveraging inherent distributions and dynamically refining soft audio-visual alignments -- probabilistic alignments between audio and visual data that capture the inherent relationships beyond explicit labels. Specifically, the model distills audio-visual distribution-based knowledge from annotated labels in a subset of each batch. This self-distilled knowledge is used t

Donghuo Zeng, Jianming Wu, Gen Hattori et al.

Learning common subspace is prevalent way in cross-modal retrieval to solve the problem of data from different modalities having inconsistent distributions and representations that cannot be directly compared. Previous cross-modal retrieval methods focus on projecting the cross-modal data into a common space by learning the correlation between them to bridge the modality gap. However, the rich semantic information in the video and the heterogeneous nature of audio-visual data leads to more serious heterogeneous gaps intuitively, which may lead to the loss of key semantic content of video with single clue by the previous methods when eliminating the modality gap, while the semantics of the categories may undermine the properties of the original features. In this work, we aim to learn effective audio-visual representations to support audio-visual cross-modal retrieval (AVCMR). We propose a novel model that maps audio-visual modalities into two distinct shared latent subspaces: explicit and implicit shared spaces. In particular, the explicit shared space is used to optimize pairwise correlations, where learned representations across modalities capture the commonalities of audio-visual pairs and reduce the modality gap. The implicit shared space is used to preserve the distinctive features between modalities by maintaining the discrimination of audio/video patterns from different semantic categories. Finally, the fusion of the features learned from the two latent subspaces is used for the similarity computation of the AVCMR task. The comprehensive experimental results on two audio-visual datasets demonstrate that our proposed model for using two different latent subspaces for audio-visual cross-modal learning is effective and significantly outperforms the state-of-the-art cross-modal models that learn features from a single subspace.

Donghuo Zeng, Jianming Wu, Bo Yang et al.

Some elderly care homes attempt to remedy the shortage of skilled caregivers and provide long-term care for the elderly residents, by enhancing the management of the care support system with the aid of smart devices such as mobile phones and tablets. Since mobile phones and tablets lack the flexibility required for laborious elderly care work, smart AR glasses have already been considered. Although lightweight smart AR devices with a transparent display are more convenient and responsive in an elderly care workplace, fetching data from the server through the Internet results in network congestion not to mention the limited display area. To devise portable smart AR devices that operate smoothly, we first present a no keep alive Internet required smart hands-free elderly care support system that employs smart glasses with facial recognition and text-to-speech synthesis technologies. Our support system utilizes automatic lightweight facial recognition to identify residents, and information about each resident in question can be obtained hands free link with a local database. Moreover, a resident information can be displayed on just a portion of the AR smart glasses on the spot. Due to the limited size of the display area, it cannot show all the necessary information. We exploit synthesized voices in the system to read out the elderly care related information. By using the support system, caregivers can gain an understanding of each resident condition immediately, instead of having to devote considerable time in advance in obtaining the complete information of all elderly residents. Our lightweight facial recognition model achieved high accuracy with fewer model parameters than current state-of-the-art methods. The validation rate of our facial recognition system was 99.3% or higher with the false accept rate of 0.001, and caregivers rated the acceptability at 3.6 (5 levels) or higher.

Donghuo Zeng, Yi Yu, Keizo Oyama

Sheet music, audio, and lyrics are three main modalities during writing a song. In this paper, we propose an unsupervised generative adversarial alignment representation (UGAAR) model to learn deep discriminative representations shared across three major musical modalities: sheet music, lyrics, and audio, where a deep neural network based architecture on three branches is jointly trained. In particular, the proposed model can transfer the strong relationship between audio and sheet music to audio-lyrics and sheet-lyrics pairs by learning the correlation in the latent shared subspace. We apply CCA components of audio and sheet music to establish new ground truth. The generative (G) model learns the correlation of two couples of transferred pairs to generate new audio-sheet pair for a fixed lyrics to challenge the discriminative (D) model. The discriminative model aims at distinguishing the input which is from the generative model or the ground truth. The two models simultaneously train in an adversarial way to enhance the ability of deep alignment representation learning. Our experimental results demonstrate the feasibility of our proposed UGAAR for alignment representation learning among sheet music, audio, and lyrics.

Donghuo Zeng

A cross-modal retrieval process is to use a query in one modality to obtain relevant data in another modality. The challenging issue of cross-modal retrieval lies in bridging the heterogeneous gap for similarity computation, which has been broadly discussed in image-text, audio-text, and video-text cross-modal multimedia data mining and retrieval. However, the gap in temporal structures of different data modalities is not well addressed due to the lack of alignment relationship between temporal cross-modal structures. Our research focuses on learning the correlation between different modalities for the task of cross-modal retrieval. We have proposed an architecture: Supervised-Deep Canonical Correlation Analysis (S-DCCA), for cross-modal retrieval. In this forum paper, we will talk about how to exploit triplet neural networks (TNN) to enhance the correlation learning for cross-modal retrieval. The experimental result shows the proposed TNN-based supervised correlation learning architecture can get the best result when the data representation extracted by supervised learning.

Donghuo Zeng, Yi Yu, Keizo Oyama

Deep learning has successfully shown excellent performance in learning joint representations between different data modalities. Unfortunately, little research focuses on cross-modal correlation learning where temporal structures of different data modalities, such as audio and video, should be taken into account. Music video retrieval by given musical audio is a natural way to search and interact with music contents. In this work, we study cross-modal music video retrieval in terms of emotion similarity. Particularly, audio of an arbitrary length is used to retrieve a longer or full-length music video. To this end, we propose a novel audio-visual embedding algorithm by Supervised Deep CanonicalCorrelation Analysis (S-DCCA) that projects audio and video into a shared space to bridge the semantic gap between audio and video. This also preserves the similarity between audio and visual contents from different videos with the same class label and the temporal structure. The contribution of our approach is mainly manifested in the two aspects: i) We propose to select top k audio chunks by attention-based Long Short-Term Memory (LSTM)model, which can represent good audio summarization with local properties. ii) We propose an end-to-end deep model for cross-modal audio-visual learning where S-DCCA is trained to learn the semantic correlation between audio and visual modalities. Due to the lack of music video dataset, we construct 10K music video dataset from YouTube 8M dataset. Some promising results such as MAP and precision-recall show that our proposed model can be applied to music video retrieval.

Haoting Liang, Donghuo Zeng, Yi Yu et al.

Since many online music services emerged in recent years so that effective music recommendation systems are desirable. Some common problems in recommendation system like feature representations, distance measure and cold start problems are also challenges for music recommendation. In this paper, I proposed a triplet neural network, exploiting both positive and negative samples to learn the representation and distance measure between users and items, to solve the recommendation task.

Donghuo Zeng, Yi Yu, Keizo Oyama

Cross-modal retrieval aims to retrieve data in one modality by a query in another modality, which has been a very interesting research issue in the field of multimedia, information retrieval, and computer vision, and database. Most existing works focus on cross-modal retrieval between text-image, text-video, and lyrics-audio.Little research addresses cross-modal retrieval between audio and video due to limited audio-video paired datasets and semantic information. The main challenge of audio-visual cross-modal retrieval task focuses on learning joint embeddings from a shared subspace for computing the similarity across different modalities, where generating new representations is to maximize the correlation between audio and visual modalities space. In this work, we propose a novel deep triplet neural network with cluster canonical correlation analysis(TNN-C-CCA), which is an end-to-end supervised learning architecture with audio branch and video branch.We not only consider the matching pairs in the common space but also compute the mismatching pairs when maximizing the correlation. In particular, two significant contributions are made: i) a better representation by constructing deep triplet neural network with triplet loss for optimal projections can be generated to maximize correlation in the shared subspace. ii) positive examples and negative examples are used in the learning stage to improve the capability of embedding learning between audio and video. Our experiment is run over 5-fold cross-validation, where average performance is applied to demonstrate the performance of audio-video cross-modal retrieval. The experimental results achieved on two different audio-visual datasets show the proposed learning architecture with two branches outperforms existing six CCA-based methods and four state-of-the-art based cross-modal retrieval methods.

Yi Yu, Samuel Beuret, Donghuo Zeng et al.

In this paper, we introduce our recent studies on human perception in audio event classification by different deep learning models. In particular, the pre-trained model VGGish is used as feature extractor to process audio data, and DenseNet is trained by and used as feature extractor for our electroencephalography (EEG) data. The correlation between audio stimuli and EEG is learned in a shared space. In the experiments, we record brain activities (EEG signals) of several subjects while they are listening to music events of 8 audio categories selected from Google AudioSet, using a 16-channel EEG headset with active electrodes. Our experimental results demonstrate that i) audio event classification can be improved by exploiting the power of human perception, and ii) the correlation between audio stimuli and EEG can be learned to complement audio event understanding.