Multi-object event graph representation learning for Video Question Answering

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.

Top-down Activity Representation Learning for Video Question Answering

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.

Parameter-Level Soft-Masking for Continual Learning

Existing research on task incremental learning in continual learning has primarily focused on preventing catastrophic forgetting (CF). Although several techniques have achieved learning with no CF, they attain it by letting each task monopolize a sub-network in a shared network, which seriously limits knowledge transfer (KT) and causes over-consumption of the network capacity, i.e., as more tasks are learned, the performance deteriorates. The goal of this paper is threefold: (1) overcoming CF, (2) encouraging KT, and (3) tackling the capacity problem. A novel technique (called SPG) is proposed that soft-masks (partially blocks) parameter updating in training based on the importance of each parameter to old tasks. Each task still uses the full network, i.e., no monopoly of any part of the network by any task, which enables maximum KT and reduction in capacity usage. To our knowledge, this is the first work that soft-masks a model at the parameter-level for continual learning. Extensive experiments demonstrate the effectiveness of SPG in achieving all three objectives. More notably, it attains significant transfer of knowledge not only among similar tasks (with shared knowledge) but also among dissimilar tasks (with little shared knowledge) while mitigating CF.

Debiasing Graph Transfer Learning via Item Semantic Clustering for Cross-Domain Recommendations

Deep learning-based recommender systems may lead to over-fitting when lacking training interaction data. This over-fitting significantly degrades recommendation performances. To address this data sparsity problem, cross-domain recommender systems (CDRSs) exploit the data from an auxiliary source domain to facilitate the recommendation on the sparse target domain. Most existing CDRSs rely on overlapping users or items to connect domains and transfer knowledge. However, matching users is an arduous task and may involve privacy issues when data comes from different companies, resulting in a limited application for the above CDRSs. Some studies develop CDRSs that require no overlapping users and items by transferring learned user interaction patterns. However, they ignore the bias in user interaction patterns between domains and hence suffer from an inferior performance compared with single-domain recommender systems. In this paper, based on the above findings, we propose a novel CDRS, namely semantic clustering enhanced debiasing graph neural recommender system (SCDGN), that requires no overlapping users and items and can handle the domain bias. More precisely, SCDGN semantically clusters items from both domains and constructs a cross-domain bipartite graph generated from item clusters and users. Then, the knowledge is transferred via this cross-domain user-cluster graph from the source to the target. Furthermore, we design a debiasing graph convolutional layer for SCDGN to extract unbiased structural knowledge from the cross-domain user-cluster graph. Our Experimental results on three public datasets and a pair of proprietary datasets verify the effectiveness of SCDGN over state-of-the-art models in terms of cross-domain recommendations.