CHA: Categorical Hierarchy-based Attention for Next POI Recommendation

Next Point-of-interest (POI) recommendation is a key task in improving location-related customer experiences and business operations, but yet remains challenging due to the substantial diversity of human activities and the sparsity of the check-in records available. To address these challenges, we proposed to explore the category hierarchy knowledge graph of POIs via an attention mechanism to learn the robust representations of POIs even when there is insufficient data. We also proposed a spatial-temporal decay LSTM and a Discrete Fourier Series-based periodic attention to better facilitate the capturing of the personalized behavior pattern. Extensive experiments on two commonly adopted real-world location-based social networks (LBSNs) datasets proved that the inclusion of the aforementioned modules helps to boost the performance of next and next new POI recommendation tasks significantly. Specifically, our model in general outperforms other state-of-the-art methods by a large margin.

Toward Automatically Labeling Situations in Soccer

We study the automatic annotation of situations in soccer games. At first sight, this translates nicely into a standard supervised learning problem. However, in a fully supervised setting, predictive accuracies are supposed to correlate positively with the amount of labeled situations: more labeled training data simply promise better performance. Unfortunately, non-trivially annotated situations in soccer games are scarce, expensive and almost always require human experts; a fully supervised approach appears infeasible. Hence, we split the problem into two parts and learn (i) a meaningful feature representation using variational autoencoders on unlabeled data at large scales and (ii) a large-margin classifier acting in this feature space but utilize only a few (manually) annotated examples of the situation of interest. We propose four different architectures of the variational autoencoder and empirically study the detection of corner kicks, crosses and counterattacks. We observe high predictive accuracies above 90% AUC irrespectively of the task.

Camera Calibration Using Catenary

In this work, we introduce a novel calibration technique based on a hanging chain curve replacing the checkerboard-based methods. It is a known physical phenomenon that a hanging chain or a flexible rope under gravity can be modeled by a special curve called catenary. Therefore, instead of the commonly-used planar calibrator, we propose using multiple shots of a catenary-shaped chain for calibration. This approach can solve the out-of-focus problem which is faced in checkerboard calibration methods when the size of the board is not large enough. Although enlarging a planar calibrator increases the manufacturing time and cost, a simple label chain can create large planar areas as precise as a rigid checkerboard, is easily foldable and transportable. We compare the results of our proposed approach against the widely used checkerboardbased calibration as well as the state-of-the-art calibration methods and show that catenary-based calibration is much more accurate than checkerboard-based calibration by a very large margin and is also very competitive among the other approaches.<br>

Camera Calibration Using Catenary

In this work, we introduce a novel calibration technique based on a hanging chain curve replacing the checkerboard-based methods. It is a known physical phenomenon that a hanging chain or a flexible rope under gravity can be modeled by a special curve called catenary. Therefore, instead of the commonly-used planar calibrator, we propose using multiple shots of a catenary-shaped chain for calibration. This approach can solve the out-of-focus problem which is faced in checkerboard calibration methods when the size of the board is not large enough. Although enlarging a planar calibrator increases the manufacturing time and cost, a simple label chain can create large planar areas as precise as a rigid checkerboard, is easily foldable and transportable. We compare the results of our proposed approach against the widely used checkerboardbased calibration as well as the state-of-the-art calibration methods and show that catenary-based calibration is much more accurate than checkerboard-based calibration by a very large margin and is also very competitive among the other approaches.<br>

Proteome-wide landscape of solubility limits in a bacterial cell

Proteins are prone to aggregate when they are expressed above their solubility limits, a phenomenon termed supersaturation. Aggregation may occur as proteins emerge from the ribosome or after they fold and accumulate in the cell, but the relative importance of these two routes remain poorly known. Here, we systematically probed the solubility limits of each Escherichia coli protein upon overexpression using an image-based screen coupled with machine learning. The analysis suggests that competition between folding and aggregation from the unfolded state governs the two aggregation routes. Remarkably, the majority (70%) of insoluble proteins have low supersaturation risks in their unfolded states and rather aggregate after folding. Furthermore, a substantial fraction (~35%) of the proteome remain soluble at concentrations much higher than those found naturally, indicating a large margin of safety to tolerate gene expression changes. We show that high disorder content and low surface stickiness are major determinants of high solubility and are favored in abundant bacterial proteins. Overall, our proteome-wide study provides empirical insights into the molecular determinants of protein aggregation routes in a bacterial cell.

Exploiting better motion cues for better action recognition

Several recent studies on action recognition have emphasised the significance of including motioncharacteristics clearly in the video description. This work shows that properly partitioning visualmotion into dominant and residual motions enhances action recognition algorithms greatly, both interms of extracting space-time trajectories and computing descriptors. Then, using differentialmotion scalar variables, divergence, curl, and shear characteristics, we create a new motiondescriptor, the DCS descriptor. It adds to the results by capturing additional information on localmotion patterns. Finally, adopting the recently proposed VLAD coding technique in image retrievalimproves action recognition significantly. On three difficult datasets, namely Hollywood 2,HMDB51, and Olympic Sports, our three additions are complementary and lead to beat all reportedresults by a large margin.

Interaction is Necessary for Distributed Learning with Privacy or Communication Constraints

Local differential privacy (LDP) is a model where users send privatized data to an untrusted central server whose goal it to solve some data analysis task. In the non-interactive version of this model the protocol consists of a single round in which a server sends requests to all users then receives their responses. This version is deployed in industry due to its practical advantages and has attracted significant research interest. Our main result is an exponential lower bound on the number of samples necessary to solve the standard task of learning a large-margin linear separator in the non-interactive LDP model. Via a standard reduction this lower bound implies an exponential lower bound for stochastic convex optimization and specifically, for learning linear models with a convex, Lipschitz and smooth loss. These results answer the questions posed by Smith, Thakurta, and Upadhyay (IEEE Symposium on Security and Privacy 2017) and Daniely and Feldman (NeurIPS 2019). Our lower bound relies on a new technique for constructing pairs of distributions with nearly matching moments but whose supports can be nearly separated by a large margin hyperplane. These lower bounds also hold in the model where communication from each user is limited and follow from a lower bound on learning using non-adaptive statistical queries.