The development of virtual reality (VR) technology has allowed virtual museums to enrich experiences of personal memories. In this work, we first conducted a survey study to understand the way people use to record, store, and share their digital memories. Informed by the results, we present MemoryVR, a personal virtual museum system designed to preserve and share digital memories. It allows individuals to organize and share their own memories in a virtual museum environment and to visit the memory spaces of others in an immersive way. We invited participants to experience MemoryVR and analyzed their behavior and expectations. The evaluation results showed that users perceived excellent pragmatic and hedonic qualities of MemoryVR. Participants found their experiences of memories in personal virtual museums to be fulfilling. Additionally, we gathered suggestions from participants about MemoryVR, providing design recommendations for customized personal virtual museums.

Identifying the hierarchical heavy hitters (HHHs), i.e., the frequent aggregated flows based on common IP prefixes, is a vital task in network traffic measurement and security. Existing methods typically employ dynamic trie structures to track numerous prefixes or utilize multiple separate sketch instances, one for each hierarchical level, to capture HHHs across different levels, while both approaches suffer from low memory efficiency and limited compatibility with programmable switches. In this paper, we introduce two novel HHH detection solutions, respectively, Hierarchical Heavy Detector (HHD) and the Compressed Hierarchical Heavy Detector (CHHD), to achieve high memory efficiency and enhanced hardware compatibility. The key idea of HHD is to design a shared bucket array structure to identify and record HHHs from all hierarchical levels, which avoids the memory wastage of maintaining separate sketches to achieve high memory efficiency and allows feasible deployment of both byte-hierarchy and bit-hierarchy HHH detection on programmable switches using minimal processing stage resources. Additionally, HHD utilizes a sampling-based update strategy to effectively balance packet processing speed and detection accuracy. Furthermore, we present the CHHD, which enhances HHH detection in bit hierarchies through a more compact cell structure, which allows for compressing several ancestor and descendant prefixes within a single cell, further boosting memory efficiency and accuracy. We have implemented HHD and CHHD on a P4-based programmable switch with limited switch resources. Experimental results based on real-world Internet traces demonstrate that HHD and CHHD outperform the state-of-the-art by achieving up to 56 percentage points higher detection precision and 2.6× higher throughput.

Out-of-distribution (OOD) detection is critical for ensuring the reliability of deep learning models in open-world applications. While post-hoc methods are favored for their efficiency and ease of deployment, existing approaches often underexploit the rich information embedded in the model’s logits space. In this paper, we propose LogitGap, a novel post-hoc OOD detection method that explicitly exploits the relationship between the maximum logit and the remaining logits to enhance the separability between in-distribution (ID) and OOD samples. To further improve its effectiveness, we refine LogitGap by focusing on a more compact and informative subset of the logit space. Specifically, we introduce a training-free strategy that automatically identifies the most informative logits for scoring. We provide both theoretical analysis and empirical evidence to validate the effectiveness of our approach. Extensive experiments on both vision-language and vision-only models demonstrate that LogitGap consistently achieves state-of-the-art performance across diverse OOD detection scenarios and benchmarks.

Pre-trained vision-language models (e.g., CLIP) have shown powerful zero-shot transfer capabilities. But they still struggle with domain shifts and typically require labeled data to adapt to downstream tasks, which could be costly. In this work, we aim to leverage unlabeled data that naturally spans multiple domains to enhance the transferability of vision-language models. Under this unsupervised multi-domain setting, we have identified inherent model bias within CLIP, notably in its visual and text encoders. Specifically, we observe that CLIP’s visual encoder tends to prioritize encoding domain over discriminative category information, meanwhile its text encoder exhibits a preference for domain-relevant classes. To mitigate this model bias, we propose a training-free and label-free feature calibration method, Unsupervised Multi-domain Feature Calibration (UMFC). UMFC estimates image-level biases from domain-specific features and text-level biases from the direction of domain transition. These biases are subsequently subtracted from original image and text features separately, to render them domain-invariant. We evaluate our method on multiple settings including transductive learning and test-time adaptation. Extensive experiments show that our method outperforms CLIP and performs on par with the state-of-the-arts that need additional annotations or optimization. Our code is available at https://github.com/GIT-LJc/UMFC.

Traditional semi-supervised learning (SSL) assumes that the feature distributions of labeled and unlabeled data are consistent which rarely holds in realistic scenarios. In this paper, we propose a novel SSL setting, where unlabeled samples are drawn from a mixed distribution that deviates from the feature distribution of labeled samples. Under this setting, previous SSL methods tend to predict wrong pseudo-labels with the model fitted on labeled data, resulting in noise accumulation. To tackle this issue, we propose \emph{Self-Supervised Feature Adaptation} (SSFA), a generic framework for improving SSL performance when labeled and unlabeled data come from different distributions. SSFA decouples the prediction of pseudo-labels from the current model to improve the quality of pseudo-labels. Particularly, SSFA incorporates a self-supervised task into the SSL framework and uses it to adapt the feature extractor of the model to the unlabeled data. In this way, the extracted features better fit the distribution of unlabeled data, thereby generating high-quality pseudo-labels. Extensive experiments show that our proposed SSFA is applicable to various pseudo-label-based SSL learners and significantly improves performance in labeled, unlabeled, and even unseen distributions.

In mobile crowd sensing (MCS), appropriate rewards are always expected to compensate the participants for their consumptions of physical resources and involvements of manual efforts. Hence, the research about incentive mechanisms is essential and useful for MCS, and the existing works focus on optimizing one of the performance, such as the data quality, or the platforms profit, so, how to design an incentive mechanism which not only can guarantee the high data quality but also can maximize the profit of platforms is challenging issue. In this paper, we propose a quality-based online incentive mechanism with unknown users cost distributions for MCS. The scheme uses a Markov Decision Procession (MDP) to online adjust the incentive price such that it tends to the optimal price. Moreover, the data quality request is also considered in our scheme. Furthermore, we evaluate the performances of our proposed scheme with single task and multiple tasks, extensive simulation results show that can perform better in profit and regret than the existing incentive scheme.