Structured sparsity enables reduced runtime orchestration on FPGAs. We present DTCore, a PC-less, compiler-directed compute engine. It consumes pre-scheduled 128-bit VLIW bundles, eliminating complex instruction fetch and dynamic arbitration. Using interleaved multi-threading (IMT) and explicit barriers, DTCore ensures deterministic synchronization. A Xilinx ZCU102 prototype closes timing at 400 MHz with fewer than 10k LUTs. Overall, offloading control to compile-time yields a remarkably lightweight, high-frequency FPGA accelerator.

Preview recommendations serve as a crucial shortcut for attracting users’ attention on various systems, platforms, and webpages, significantly boosting user engagement. However, the variability of preview types and the flexibility of preview duration make it challenging to use an integrated framework for multi-preview recommendations under resource constraints. In this paper, we present an approach that incorporates constrained Q-learning into a notification recommendation system, effectively handling both multi-preview ranking and duration orchestration by targeting long-term user retention. Our method bridges the gap between combinatorial reinforcement learning, which often remains too theoretical for practical use, and segmented modules in production, where model performance is typically compromised due to over-simplification. We demonstrate the superiority of our approach through off-policy evaluation and online A/B testing using Microsoft data.

Retinal blood vessels structure analysis is an important step in the detection of ocular diseases such as diabetic retinopathy and retinopathy of prematurity. Accurate tracking and estimation of retinal blood vessels in terms of their diameter remains a major challenge in retinal structure analysis. In this research, we develop a rider-based Gaussian approach for accurate tracking and diameter estimation of retinal blood vessels. The diameter and curvature of the blood vessel are assumed as the Gaussian processes. The features are determined for training the Gaussian process using Radon transform. The kernel hyperparameter of Gaussian processes is optimized using Rider Optimization Algorithm for evaluating the direction of the vessel. Multiple Gaussian processes are used for detecting the bifurcations and the difference in the prediction direction is quantified. The performance of the proposed Rider-based Gaussian process is evaluated with mean and standard deviation. Our method achieved high performance with the standard deviation of 0.2499 and mean average of 0.0147, which outperformed the state-of-the-art method by 6.32%. Although the proposed model outperformed the state-of-the-art method in normal blood vessels, in future research, one can include tortuous blood vessels of different retinopathy patients, which would be more challenging due to large angle variations. We used Rider-based Gaussian process for tracking blood vessels to obtain the diameter of retinal blood vessels, and the method performed well on the "STrutred Analysis of the REtina (STARE) Database" accessed on Oct. 2020 (https://cecas.clemson.edu/~ahoover/stare/). To the best of our knowledge, this experiment is one of the most recent analysis using this type of algorithm.

Hashtag is an important advertising tool and a must-have feature for social media nowadays. In the past, many hashtag recommendation methods have been proposed from different perspectives of images, texts, and users. However, most previous works consider neither the mutual influence between multi-modalities, nor the visual similarity between images. In this paper, we devise a novel model, namedTriplet-Attention Graph Network(TAGNet). The rationale behind our method is that visually similar images share some common hashtags. Therefore, we build an image graph, and apply a newAggregated Graph Convolution(AGC) module to propagate information in a collective way. Furthermore, it is noted that text and user also have rich content information within posts, and we hence propose aTriplet Attention(TA) module to incorporate multi-modalities into node features. Experiments on the large-scale dataset collected from Instagram show that TAGNet achieved significant improvement in recall rate over the best state-of-the-art method.

Mobile phones have become a necessity of modern life and are used for various tasks, including audio calls, text messaging, Web surfing, video streaming, gaming, and payments. Most of these tasks consume significant battery power, which makes public charging services increasingly important. Therefore, there are many public places and private stores have started to provide free charging services. However, there are several issues that restrict the popularity of public charging service. First, people cannot easily find charging spots. Second, there is no effective way to monitor charging status and manage the chargers. Third, a free charging service increases business expenses. To address these issues, we developed a Internet of Things (IoT)-based wireless charging service system, which is called AnyCharge. There are five major components of AnyCharge: 1) a Wi-Fi enabled wireless charger; 2) an IoT gateway; 3) a cloud-based management platform; 4) a secure Wi-Fi auto-connection algorithm; and 5) a mobile app. The charger is connected to an IoT gateway through Wi-Fi using our secure auto-connection algorithm, and the gateways are linked to the cloud server using message queue telemetry transport. The administrators can monitor and control chargers using the management platform. In addition, Android and iOS apps have been created to allow users to locate free chargers and find the shortest route to the nearest charging spot. We initiated a large-scale experimental deployment in Taiwan, Thailand, Singapore, and Japan. More than 200 chargers have been installed in different places, including restaurants, hospitals, telecom stores, and hotels. Statistics shows that there is a strong demand for public charging services.

Due to the advances in deep reinforcement learning and the demand of large training data, virtual-to-real learning has gained lots of attention from computer vision community recently. As state-of-the-art 3D engines can generate photo-realistic images suitable for training deep neural networks, researchers have been gradually applied 3D virtual environment to learn different tasks including autonomous driving, collision avoidance, and image segmentation, to name a few. Although there are already many open-source simulation environments readily available, most of them either provide small scenes or have limited interactions with objects in the environment. To facilitate visual recognition learning, we present a new Virtual Environment for Visual Deep Learning (VIVID), which offers large-scale diversified indoor and outdoor scenes. Moreover, VIVID leverages the advanced human skeleton system, which enables us to simulate numerous complex human actions. VIVID has a wide range of applications and can be used for learning indoor navigation, action recognition, event detection, etc. We also release several deep learning examples in Python to demonstrate the capabilities and advantages of our system.

Telecommunication fraud is one of the most prevalent crimes nowadays, and causes most property loss of victims. The criminals of telecommunication fraud are highly organized, concealed and transnational, making investigators difficult to track and capture the suspects. In this paper, we propose a Telecom Fraud Analysis Model (TFAM) which can unveil the underlying structure of fraud groups and identify the roles of the fraudsters. The links between suspects are built using flight information, and co-offending records. Social network analysis techniques are applied to analyze group structures as well as influences of each member. We collect a real telecom fraud dataset with 113 fraudsters whose fraudulent activities spread across four countries and 17 cities. Experimental results demonstrate that our method can successfully identify the key roles and discover the hidden structure of the fraud groups.

Late fusion is one of the most effective approaches to enhance recognition accuracy through combining prediction scores of multiple classifiers, each of which is trained by a specific feature or model. The existing methods generally use a fixed fusion weight for one classifier over all samples, and ignore the fact that each classifier may perform better or worse for different subsets of samples. In order to address this issue, we propose a novel sample specific late fusion (SSLF) method. Specifically, we cast late fusion into an information propagation process that diffuses the fusion weights of labeled samples to the individual unlabeled samples, and enforce positive samples to have higher fusion scores than negative samples. Upon this process, the optimal fusion weight for each sample is identified, while positive samples are pushed toward the top at the fusion score rank list to achieve better accuracy. In this paper, two SSLF methods are presented. The first method is ranking SSLF (R-SSLF), which is based on graph Laplacian with RankSVM style constraints. We formulate and solve the problem with a fast gradient projection algorithm; the second method is infinite push SSLF (I-SSLF), which combines graph Laplacian with infinite push constraints. I-SSLF is a l∞ norm constrained optimization problem and can be solved by an efficient alternating direction method of multipliers method. Extensive experiments on both large-scale image and video data sets demonstrate the effectiveness of our methods. In addition, in order to make our method scalable to support large data sets, the AnchorGraph model is employed to propagate information on a subset of samples (anchor points) and then reconstruct the entire graph to get the weights of all samples. To the best of our knowledge, this is the first method that supports learning of sample specific fusion weights for late fusion.

Video event detection allows intelligent indexing of video content based on events. Traditional approaches extract features from video frames or shots, then quantize and pool the features to form a single vector representation for the entire video. Though simple and efficient, the final pooling step may lead to loss of temporally local information, which is important in indicating which part in a long video signifies presence of the event. In this work, we propose a novel instance-based video event detection approach. We represent each video as multiple 'instances', defined as video segments of different temporal intervals. The objective is to learn an instance-level event detection model based on only video-level labels. To solve this problem, we propose a large-margin formulation which treats the instance labels as hidden latent variables, and simultaneously infers the instance labels as well as the instance-level classification model. Our framework infers optimal solutions that assume positive videos have a large number of positive instances while negative videos have the fewest ones. Extensive experiments on large-scale video event datasets demonstrate significant performance gains. The proposed method is also useful in explaining the detection results by localizing the temporal segments in a video which is responsible for the positive detection.

Late fusion addresses the problem of combining the prediction scores of multiple classifiers, in which each score is predicted by a classifier trained with a specific feature. However, the existing methods generally use a fixed fusion weight for all the scores of a classifier, and thus fail to optimally determine the fusion weight for the individual samples. In this paper, we propose a sample-specific late fusion method to address this issue. Specifically, we cast the problem into an information propagation process which propagates the fusion weights learned on the labeled samples to individual unlabeled samples, while enforcing that positive samples have higher fusion scores than negative samples. In this process, we identify the optimal fusion weights for each sample and push positive samples to top positions in the fusion score rank list. We formulate our problem as a L∞norm constrained optimization problem and apply the Alternating Direction Method of Multipliers for the optimization. Extensive experiment results on various visual categorization tasks show that the proposed method consistently and significantly beats the state-of-the-art late fusion methods. To the best knowledge, this is the first method supporting sample-specific fusion weight learning.