Vision-Language Pre-training (VLP) shows remarkable progress with the assistance of extremely heavy parameters, which challenges deployment in real applications. Knowledge distillation is well recognized as the essential procedure in model compression. However, existing knowledge distillation techniques lack an in-depth investigation and analysis of VLP, and practical guidelines for VLP-oriented distillation are still not yet explored. In this paper, we present DLIP, a simple yet efficient Distilling Language-Image Pre-training framework, through which we investigate how to distill a light VLP model. Specifically, we dissect the model distillation from multiple dimensions, such as the architecture characteristics of different modules and the information transfer of different modalities. We conduct comprehensive experiments and provide insights on distilling a light but performant VLP model. Experimental results reveal that DLIP can achieve a state-of-the-art accuracy/efficiency trade-off across diverse cross-modal tasks, e.g., image-text retrieval, image captioning and visual question answering. For example, DLIP compresses BLIP by 1.9x, from 213M to 108M parameters, while achieving comparable or better performance. Furthermore, DLIP succeeds in retaining more than 95% of the performance with 22.4% parameters and 24.8% FLOPs compared to the teacher model and accelerates inference speed by 2.7x.
Multimodal-aware recommender systems (MRSs) exploit multimodal content (e.g., product images or descriptions) as items' side information to improve recommendation accuracy. While most of such methods rely on factorization models (e.g., MFBPR) as base architecture, it has been shown that MFBPR may be affected by popularity bias, meaning that it inherently tends to boost the recommendation of popular (i.e., short-head) items at the detriment of niche (i.e., long-tail) items from the catalog. Motivated by this assumption, in this work, we provide one of the first analyses on how multimodality in recommendation could further amplify popularity bias. Concretely, we evaluate the performance of four state-of-the-art MRSs algorithms (i.e., VBPR, MMGCN, GRCN, LATTICE) on three datasets from Amazon by assessing, along with recommendation accuracy metrics, performance measures accounting for the diversity of recommended items and the portion of retrieved niche items. To better investigate this aspect, we decide to study the separate influence of each modality (i.e., visual and textual) on popularity bias in different evaluation dimensions. Results, which demonstrate how the single modality may augment the negative effect of popularity bias, shed light on the importance to provide a more rigorous analysis of the performance of such models.
Emotion recognition in conversations (ERC) is a crucial task for building human-like conversational agents. While substantial efforts have been devoted to ERC for chit-chat dialogues, the task-oriented counterpart is largely left unattended. Directly applying chit-chat ERC models to task-oriented dialogues (ToDs) results in suboptimal performance as these models overlook key features such as the correlation between emotions and task completion in ToDs. In this paper, we propose a framework that turns a chit-chat ERC model into a task-oriented one, addressing three critical aspects: data, features and objective. First, we devise two ways of augmenting rare emotions to improve ERC performance. Second, we use dialogue states as auxiliary features to incorporate key information from the goal of the user. Lastly, we leverage a multi-aspect emotion definition in ToDs to devise a multi-task learning objective and a novel emotion-distance weighted loss function. Our framework yields significant improvements for a range of chit-chat ERC models on EmoWOZ, a large-scale dataset for user emotion in ToDs. We further investigate the generalisability of the best resulting model to predict user satisfaction in different ToD datasets. A comparison with supervised baselines shows a strong zero-shot capability, highlighting the potential usage of our framework in wider scenarios.
Magnetic resonance angiography (MRA) is an imaging modality for visualising blood vessels. It is useful for several diagnostic applications and for assessing the risk of adverse events such as haemorrhagic stroke (resulting from the rupture of aneurysms in blood vessels). However, MRAs are not acquired routinely, hence, an approach to synthesise blood vessel segmentations from more routinely acquired MR contrasts such as T1 and T2, would be useful. We present an encoder-decoder model for synthesising segmentations of the main cerebral arteries in the circle of Willis (CoW) from only T2 MRI. We propose a two-phase multi-objective learning approach, which captures both global and local features. It uses learned local attention maps generated by dilating the segmentation labels, which forces the network to only extract information from the T2 MRI relevant to synthesising the CoW. Our synthetic vessel segmentations generated from only T2 MRI achieved a mean Dice score of $0.79 \pm 0.03$ in testing, compared to state-of-the-art segmentation networks such as transformer U-Net ($0.71 \pm 0.04$) and nnU-net($0.68 \pm 0.05$), while using only a fraction of the parameters. The main qualitative difference between our synthetic vessel segmentations and the comparative models was in the sharper resolution of the CoW vessel segments, especially in the posterior circulation.
Graph Convolutional Networks (GCNs) can capture non-Euclidean spatial dependence between different brain regions, and the graph pooling operator in GCNs is key to enhancing the representation learning capability and acquiring abnormal brain maps. However, the majority of existing research designs graph pooling operators only from the perspective of nodes while disregarding the original edge features, in a way that not only confines graph pooling application scenarios, but also diminishes its ability to capture critical substructures. In this study, a clustering graph pooling method that first supports multidimensional edge features, called Edge-aware hard clustering graph pooling (EHCPool), is developed. EHCPool proposes the first 'Edge-to-node' score evaluation criterion based on edge features to assess node feature significance. To more effectively capture the critical subgraphs, a novel Iteration n-top strategy is further designed to adaptively learn sparse hard clustering assignments for graphs. Subsequently, an innovative N-E Aggregation strategy is presented to aggregate node and edge feature information in each independent subgraph. The proposed model was evaluated on multi-site brain imaging public datasets and yielded state-of-the-art performance. We believe this method is the first deep learning tool with the potential to probe different types of abnormal functional brain networks from data-driven perspective.
Surgical context inference has recently garnered significant attention in robot-assisted surgery as it can facilitate workflow analysis, skill assessment, and error detection. However, runtime context inference is challenging since it requires timely and accurate detection of the interactions among the tools and objects in the surgical scene based on the segmentation of video data. On the other hand, existing state-of-the-art video segmentation methods are often biased against infrequent classes and fail to provide temporal consistency for segmented masks. This can negatively impact the context inference and accurate detection of critical states. In this study, we propose a solution to these challenges using a Space Time Correspondence Network (STCN). STCN is a memory network that performs binary segmentation and minimizes the effects of class imbalance. The use of a memory bank in STCN allows for the utilization of past image and segmentation information, thereby ensuring consistency of the masks. Our experiments using the publicly available JIGSAWS dataset demonstrate that STCN achieves superior segmentation performance for objects that are difficult to segment, such as needle and thread, and improves context inference compared to the state-of-the-art. We also demonstrate that segmentation and context inference can be performed at runtime without compromising performance.
The analysis and mining of user heterogeneous behavior are of paramount importance in recommendation systems. However, the conventional approach of incorporating various types of heterogeneous behavior into recommendation models leads to feature sparsity and knowledge fragmentation issues. To address this challenge, we propose a novel approach for personalized recommendation via Large Language Model (LLM), by extracting and fusing heterogeneous knowledge from user heterogeneous behavior information. In addition, by combining heterogeneous knowledge and recommendation tasks, instruction tuning is performed on LLM for personalized recommendations. The experimental results demonstrate that our method can effectively integrate user heterogeneous behavior and significantly improve recommendation performance.
Recently, inference privacy has attracted increasing attention. The inference privacy concern arises most notably in the widely deployed edge-cloud video analytics systems, where the cloud needs the videos captured from the edge. The video data can contain sensitive information and subject to attack when they are transmitted to the cloud for inference. Many privacy protection schemes have been proposed. Yet, the performance of a scheme needs to be determined by experiments or inferred by analyzing the specific case. In this paper, we propose a new metric, \textit{privacy protectability}, to characterize to what degree a video stream can be protected given a certain video analytics task. Such a metric has strong operational meaning. For example, low protectability means that it may be necessary to set up an overall secure environment. We can also evaluate a privacy protection scheme, e.g., assume it obfuscates the video data, what level of protection this scheme has achieved after obfuscation. Our definition of privacy protectability is rooted in information theory and we develop efficient algorithms to estimate the metric. We use experiments on real data to validate that our metric is consistent with empirical measurements on how well a video stream can be protected for a video analytics task.
We design a Universal Automatic Elbow Detector (UAED) for deciding the effective number of components in model selection problems. The relationship with the information criteria widely employed in the literature is also discussed. The proposed UAED does not require the knowledge of a likelihood function and can be easily applied in diverse applications, such as regression and classification, feature and/or order selection, clustering, and dimension reduction. Several experiments involving synthetic and real data show the advantages of the proposed scheme with benchmark techniques in the literature.
Current closed-set instance segmentation models rely on pre-defined class labels for each mask during training and evaluation, largely limiting their ability to detect novel objects. Open-world instance segmentation (OWIS) models address this challenge by detecting unknown objects in a class-agnostic manner. However, previous OWIS approaches completely erase category information during training to keep the model's ability to generalize to unknown objects. In this work, we propose a novel training mechanism termed SegPrompt that uses category information to improve the model's class-agnostic segmentation ability for both known and unknown categories. In addition, the previous OWIS training setting exposes the unknown classes to the training set and brings information leakage, which is unreasonable in the real world. Therefore, we provide a new open-world benchmark closer to a real-world scenario by dividing the dataset classes into known-seen-unseen parts. For the first time, we focus on the model's ability to discover objects that never appear in the training set images. Experiments show that SegPrompt can improve the overall and unseen detection performance by 5.6% and 6.1% in AR on our new benchmark without affecting the inference efficiency. We further demonstrate the effectiveness of our method on existing cross-dataset transfer and strongly supervised settings, leading to 5.5% and 12.3% relative improvement.