Automatic emotion recognition has recently gained significant attention due to the growing popularity of deep learning algorithms. One of the primary challenges in emotion recognition is effectively utilizing the various cues (modalities) available in the data. Another challenge is providing a proper explanation of the outcome of the learning. To address these challenges, we present Explainable Multimodal Emotion Recognition with Situational Knowledge (EMERSK), a generalized and modular system for human emotion recognition and explanation using visual information. Our system can handle multiple modalities, including facial expressions, posture, and gait, in a flexible and modular manner. The network consists of different modules that can be added or removed depending on the available data. We utilize a two-stream network architecture with convolutional neural networks (CNNs) and encoder-decoder style attention mechanisms to extract deep features from face images. Similarly, CNNs and recurrent neural networks (RNNs) with Long Short-term Memory (LSTM) are employed to extract features from posture and gait data. We also incorporate deep features from the background as contextual information for the learning process. The deep features from each module are fused using an early fusion network. Furthermore, we leverage situational knowledge derived from the location type and adjective-noun pair (ANP) extracted from the scene, as well as the spatio-temporal average distribution of emotions, to generate explanations. Ablation studies demonstrate that each sub-network can independently perform emotion recognition, and combining them in a multimodal approach significantly improves overall recognition performance. Extensive experiments conducted on various benchmark datasets, including GroupWalk, validate the superior performance of our approach compared to other state-of-the-art methods.

Multipath TCP (MPTCP) enables a client to exploit multiple wireless paths simultaneously for improved throughput and mobility resilience. However, MPTCP clients in WiFi networks may easily lower the achievable throughput of the network due to excessively created connections over access points (APs), particularly those that own a weak link quality. This is caused by the fact that the maximal achievable throughput of a WiFi AP can be easily affected by the number of connections accessing it and the link qualities of those connections. In this paper, we first verify such effects through extensive experiments and analysis. We then propose a novel scheme, denoted MPWiFi, to solve this issue while keeping the benefits of MPTCP based multipath access in WiFi networks. Our scheme allows a MPTCP client to obtain resources on its best WiFi path freely and suppresses its subflows in additional paths when congestion happens. As a result, the degrading to the WiFi AP's achievable throughput is greatly prevented with acceptable influences on the benefits of MPTCP clients (i.e., their multipath access is intervened only on congested non-best APs). The fairness to MPTCP clients is also guaranteed through their best WiFi connection. The proposed solution is implemented along with the Linux Kernel MPTCP implementation. Extensive real-world deployment based experiments and NS3 simulation show that the proposed scheme can effectively alleviate the adverse impact of MPTCP based multipath access in WiFi networks while keeping its benefits.

Correlating the congestion control (CC) of parallel subflows of multipath TCP (MPTCP) has shown the advantage of making it fair and friendly to legacy TCP. But the correlation also leads to some new drawbacks. In this paper, we first analyze major correlated MPTCP CC algorithms through the perspective of bandwidth competition. Based on the modeling, we verify, discover, and explain three shortages of correlated CC, namely 1) limit fairness semantic; 2) render new attack surfaces; and 3) interplay with network sharing policies. We further find that decoupling the CC of subflows is more promising in solving the above issues. However, directly applying legacy TCP CC to each subflow independently would lose the benefits of correlated CC. This motivates the design of an algorithm that decouples the CC of subflows as much as possible while still making MPTCP fair and friendly. To attain this goal, we translate the goals of correlated CC into an approximation principle under the decoupled semantic. We then propose a self-Constrained And Decoupled Increase Algorithm (CADIA) that achieves the principle by adaptively detecting and constraining non-best subflows. This makes CADIA share the benefits of correlated CC algorithms, while owning no direct correlation among the CC of subflows. Extensive analysis and experiments are conducted to demonstrate the effectiveness of CADIA.

The emergence of multipath TCP (MPTCP) has made it much easier for mobile devices to use multiple wireless network access paths simultaneously. However, we find that a large amount of multipath users could negatively impact the performance of wireless networks in terms of throughput and fairness due to increased amount of wireless connections. Therefore, in this paper, we aim to mitigate such a problem while retaining the benefits of multipath access in wireless networks from the perspective of network owner. We solve the challenge by 1) identifying a solution principle that can effectively balance the two goals and 2) implementing the principle through an SDN based bandwidth usage management system on the network side. When there is congestion on an AP, our method enforces necessary suppressions on non-primary subflows from multipath users to protect the network performance, while keeping the multipath benefits as much as possible. When an AP owns idled capacity, no intervention is imposed, thus offering the maximal benefits to multipath users without substantially affecting the network performance. Thus, the multipath access is dynamically scaled to achieve a balance of the two goals. Extensive NS3 experiment with Linux Kernel MPTCP implementation demonstrates our findings as well as the effectiveness of the proposed system.

We found that directly employing MPTCP in WiFi network might adversely affect the overall network performance. The connecting MPTCP client achieves little bandwidth advantage through redundant paths of weak signal strength. However, it impairs the overall network performance by unintentionally reducing the throughput available to other clients sharing that access point in much greater extent. In this paper, we analyzed this phenomenon and provided a method, which improves the adverse impact of MPTCP significantly.