Recent proprietary video generation models have demonstrated remarkable proficiency in synthesizing highly realistic videos from textual instructions.Most open-source text-tovideo models, however, still struggle to accurately simulate real-world physics and dynamic entity interactions.Existing approaches rely on scaling laws and large-scale, high-quality video datasets to implicitly learn physical dynamics, yet this paradigm is constrained by prohibitive costs and the burdensome demands of data curation.Motivated by this, we propose a novel framework that integrates graph-structured temporal knowledge into video latent diffusion models to enhance compositional generation and interaction fidelity.Our framework constructs video scene graphs specifically designed to capture entity relationships, temporal dynamics, and global scene context.These graphstructured representations guide the generation process through cross-attention mechanisms.Additionally, we introduce Graph-Aligned Denoising Loss (GADL), a training objective that ensures adherence to conditioned graphs by incorporating node modification tasks within the denoising process, leveraging synchronized edited video-graph pairs.Comprehensive evaluations demonstrate that incorporating graphstructured knowledge significantly enhances compositionality and the accurate portrayal of real-world interactions in generated videos.

Learning generalized models from biased data is an important undertaking toward fairness in deep learning. To address this issue, recent studies attempt to identify and leverage bias-conflicting samples free from spurious correlations without prior knowledge of bias or an unbiased set. However, spurious correlation remains an ongoing challenge, primarily due to the difficulty in correctly detecting these samples. In this paper, inspired by the similarities between mislabeled samples and bias-conflicting samples, we approach this challenge from a novel perspective of mislabeled sample detection. Specifically, we delve into Influence Function, one of the standard methods for mislabeled sample detection, for identifying bias-conflicting samples and propose a simple yet effective remedy for biased models by leveraging them. Through comprehensive analysis and experiments on diverse datasets, we demonstrate that our new perspective can boost the precision of detection and rectify biased models effectively. Furthermore, our approach is complementary to existing methods, showing performance improvement even when applied to models that have already undergone recent debiasing techniques.

A cut-and-paste type of data augmentation strategy has attracted considerable attention in the vision community due to its simplicity and effectiveness in improving generalization performance. However, it is challenging for Automatic Speech Recognition (ASR) tasks to apply this type of augmentation since segments corresponding to specific output tokens (e.g. words or sub-words) have various lengths. Furthermore, if speech signals are indiscriminately mixed without considering semantics, the risk of generating nonsensical sentences arises. To address these issues, in this paper, we propose WeavSpeech, still a simple yet effective cut-and-paste augmentation method for ASR tasks that weaves a pair of speech data considering semantics. Our method can be applied to any language without requiring language-specific knowledge and seamlessly integrated with other verified augmentations. We validate the superiority of our method on representative ASR benchmark datasets, including LibriSpeech and WSJ.

In real-world graphs, noisy connections are inevitable, which makes it difficult to obtain unbiased node representations. Among various attempts to resolve this problem, a method of estimating the counterfactual effects of these connectivities has recently attracted attention, which mainly uses influence functions for single graph elements (i.e., node and edge). However, in this paper, we argue that there is a strongly interacting group effect between the influences of graph elements due to their connectivity. In the same vein, we observe that edge groups connecting to the same train node exhibit significant differences in their influences, hence no matter how negative each is, removing them at once may have a rather negative effect as a group. Based on this motivation, we propose a new edge-removing strategy, Repulsive edge Group Elimination (RGE), that preferentially removes edges with no interference in groups. Empirically, we demonstrate that RGE consistently outperforms existing methods on the various benchmark datasets.

Learning unbiased node representations under class-imbalanced graph data is challenging due to interactions between adjacent nodes. Existing studies have in common that they compensate the minor class nodes ‘as a group’ according to their overall quantity (ignoring node connections in graph), which inevitably increase the false positive cases for major nodes. We hypothesize that the increase in these false positive cases is highly affected by the label distribution around each node and confirm it experimentally. In addition, in order to handle this issue, we propose Topology-Aware Margin (TAM) to reflect local topology on the learning objective. Our method compares the connectivity pattern of each node with the class-averaged counter-part and adaptively adjusts the margin accordingly based on that. Our method consistently exhibits superiority over the baselines on various node classification benchmark datasets with representative GNN architectures.

Monotonic Multihead Attention, which allows multiple heads to learn their own alignments per head, shows great performance on simultaneous machine translation and streaming speech recognition. However, it causes high latency waiting for the slowest head. Some recent advances such as Head-Synchronous Beam Search Decoding and its learnable version Mutually-Constrained Monotonic Multihead Attention, try to address this issue by restricting the difference in times of chosen frames among multi-heads to a fixed waiting time threshold. In this paper, we hypothesis that the optimal threshold for high performance with low latency depends on the input sequence, and propose an adaptive algorithm that learns how long to wait depending on input tokens by introducing a threshold prediction module. We evaluate our approach on two benchmark datasets for online Automatic Speech Recognition task and demonstrate that our method reduces the latency together with even improving the recognition accuracy.

Despite the feature of real-time decoding, Monotonic Multi-head Attention (MMA) shows comparable performance to the state-of-the-art offline methods in machine translation and automatic speech recognition (ASR) tasks. However, the latency of MMA is still a major issue in ASR and should be combined with a technique that can reduce the test latency at inference time, such as head-synchronous beam search decoding, which forces all non-activated heads to activate after a small fixed delay from the first head activation. In this paper, we remove the discrepancy between training and test phases by considering, in the training of MMA, the interactions across multiple heads that will occur in the test time. Specifically, we derive the expected alignments from monotonic attention by considering the boundaries of other heads and reflect them in the learning process. We validate our proposed method on the two standard benchmark datasets for ASR and show that our approach, MMA with the mutually-constrained heads from the training stage, provides better performance than baselines.