Time Granularity Research Articles

A Novel Spatial–Temporal Deep Learning Method for Metro Flow Prediction Considering External Factors and Periodicity

Predicting metro traffic flow is crucial for efficient urban planning and transit management. It enables cities to optimize resource allocation, reduce congestion, and enhance the overall commuter experience in rapidly urbanizing environments. Nevertheless, metro flow prediction is challenging due to the intricate spatial–temporal relationships inherent in the data and the varying influence of external factors. To model spatial–temporal correlations considering external factors, a novel spatial–temporal deep learning framework is proposed in this study. Firstly, mutual information is utilized to select the highly corrected stations of the examined station. Compared with the traditional correlation calculation methods, mutual information is particularly advantageous for analyzing nonlinear metro flow data. Secondly, metro flow data reflecting the historical trends from different time granularities are incorporated. Additionally, the external factor data that influence the metro flow are also considered. Finally, these multiple sources and dimensions of data are combined and fed into the deep neural network to capture the complex correlations of multi-dimensional data. Sufficient experiments are designed and conducted on the real dataset collected from Xi’an subway to verify the effectiveness of the proposed model. Experimental results are comprehensively analyzed according to the POI information around the subway station.

Short-term trajectory prediction for individual metro passengers based on multi-level periodicity mining from semantic trajectory

Short-term trajectory prediction for individual metro passengers is an essential component of many location-based applications, such as bottleneck identification, anomalous trajectory detection, and emergency evacuation. Accurate and interpretable trajectory prediction in short-term remains challenging since the trajectory of individual metro passengers is influenced by time-varying and periodic mobility patterns. Moreover, it is still difficult for deep learning-based prediction approaches to intuitively explain the individual periodic mobility patterns captured by the black-box model. In this study, a long short-term memory method based on multi-level periodicity mining (LSTM-MP) from semantic trajectory is developed to achieve accurate and interpretable short-term trajectory prediction for individual metro passengers. The newly defined semantic trajectories of metro passengers that combine the semantic information about the geometric and operational characteristics of the metro with the raw trajectory data are constructed to help obtain fine-grained periodic mobility patterns of individual passengers. Multi-level periodicity which includes long-range, weekly-interval, and daily-interval periodicities, is calculated with variable time granularity and its polymerization from the constructed semantic trajectories. Finally, the characteristics of multi-level periodicity are considered by developing a principal trajectory attention mechanism to select the most relevant trajectories for prediction in the LSTM-MP framework. The prediction performance of LSTM-MP is compared with 2-order Markov chain and Naïve-LSTM algorithms using the large real-world anonymous passenger WiFi probe dataset. Experimental results show that LSTM-MP can predict when and where an individual will be in the future time series with an average accuracy of 82.3%. This demonstrates the effectiveness of LSTM-MP in the short-term trajectory prediction for individual metro passengers. Potential applications of the short-term trajectory prediction for individual metro passengers are also identified.

Prediction of Energy Consumption in Horizontal Roughing Process of Hot Rolling Strip Based on TDADE Algorithm

The steel industry is the key industry of energy consumption. The optimization of rolling process parameters is an effective measure to reduce and optimize energy consumption. Accurate energy consumption prediction has an indispensable guiding function in the planning of process parameters. However, the traditional energy consumption prediction and machine learning methods have been unable to fit the needs of high precision and reliability. Taking the horizontal roughing process (HRP) of hot rolling process as the research object, this paper mainly introduces an energy consumption prediction mechanism model (ECPMM) based on a roller adaptive wear strategy by analyzing the strip forming mechanism and rolling process. Then, two directions adaptive differential evolution (TDADE) algorithm is proposed based on a novel adaptive strategy. This algorithm has better convergence speed and global optimization ability than other optimization algorithms in the parameter optimization of ECPMM. We carried out comparative experiments on five machine learning algorithms. The results show that the ECPMM optimized by TDADE is superior to the five machine learning algorithms in prediction accuracy and stability. Finally, we conducted ablation experiments on the energy consumption characteristics of the HRP process, which turns out that using a smaller reduction and roller radius can reduce energy consumption. Summarizing the above experimental results suggests that the ECPMM has high accuracy and robustness, and satisfies the requirements of practical application. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —It is necessary to carry out the deep analysis and research of energy consumption prediction involved in the process planning of strip hot rolling, the inspiration of this article is stem from this point. The commonly used energy consumption prediction methods include mechanism modeling and machine learning, but both have disadvantages, like low prediction accuracy and black box, which must have to be performed to solve by adapting the production demand of high reliability. In this paper, the combination of the mechanism model and data-driven method is used to realize the accurate prediction of energy consumption and improve the reliability of the model. To the best of our knowledge, this is the first paper about the research on energy consumption prediction of hot rolling process based on mechanism model and data-driven, especially in data size, time granularity, algorithm design, and prediction performance. Such research is helpful to guide engineers to design rolling energy consumption. We will dive a bit deeper into learning the energy consumption and process parameters during the rolling process in future studies.

Multi-Spatio-Temporal Convolutional Neural Network for Short-Term Metro Passenger Flow Prediction

Accurate short-term prediction of metro passenger flow can offer significant assistance in optimizing train schedules, reducing congestion during peak times, and improving the service level of the metro system. Currently, most models do not fully utilize the high-resolution data aggregated by automatic fare collection systems. Therefore, we propose a model, called MST-GRT, that aggregates multi-time granularity data and considers multi-graph structures. Firstly, we analyze the correlation between metro passenger flow sequences at different time granularities and establish a principle for extracting the spatiotemporal correlation of data at different time granularities using the multi-graph neural network. Subsequently, we use residual blocks to construct a deep convolutional neural network to aggregate the data of different time granularities from small to large, obtaining multi-channel feature maps of multi-time granularity. To process the multi-channel feature maps, we use 2D dilated causal convolution to reconstruct the TCN (Temporal Convolutional Network) to compress the channel number of the feature maps and extract the time dependency of the data, and finally output the results through a fully connected layer. The experimental results demonstrate that our model outperforms the baseline models on the Hangzhou Metro smart-card data set.

Temporal group interaction density in collaborative problem solving: Exploring group interactions with different time granularities

Temporal group interaction density in collaborative problem solving: Exploring group interactions with different time granularities

LinoSPAD2: an FPGA-based, hardware-reconfigurable 512×1 single-photon camera system.

We report on LinoSPAD2, a single-photon camera system, comprising a 512×1 single-photon avalanche diode (SPAD) front-end and one or two FPGA-based back-ends. Digital signals generated by the SPADs are processed by the FPGA in real time, whereas the FPGA offers full reconfigurability at a very high level of granularity both in time and space domains. The LinoSPAD2 camera system can process 512 SPADs simultaneously through 256 channels, duplicated on each FPGA-based back-end, with a bank of 64 time-to-digital converters (TDCs) operating at 133 MSa/s, whereas each TDC has a time resolution of 20 ps (LSB). To the best of our knowledge, LinoSPAD2 is the first fully reconfigurable SPAD camera system of large format. The SPAD front-end features a pitch of 26.2 μm, a native fill factor of 25.1%, and a microlens array achieving 2.3× concentration factor. At room temperature, the median dark count rate (DCR) is 80 cps at 7 V excess bias, the peak photon detection probability (PDP) is 53% at 520 nm wavelength, and the single-photon timing resolution (SPTR) is 50 ps FWHM. The instrument response function (IRF) is around 100 ps FWHM at system level. The LinoSPAD2 camera system is suitable for numerous applications, including LiDAR imaging, heralded spectroscopy, compressive Raman sensing, and other computational imaging techniques.

The Building Data Genome Directory – An open, comprehensive data sharing platform for building performance research

The building sector plays a crucial role in the worldwide decarbonization effort, accounting for significant portions of energy consumption and environmental effects. However, the scarcity of open data sources is a continuous challenge for built environment researchers and practitioners. Although several efforts have been made to consolidate existing open datasets, no database currently offers a comprehensive collection of building data types with all subcategories and time granularities (e.g., year, month, and sub-hour). This paper presents the Building Data Genome Directory, an open data-sharing platform serving as a one-stop shop for the data necessary for vital categories of building energy research. The data directory is an online portal (buildingdatadirectory.org/) that allows filtering and discovering valuable datasets. The directory covers meter, building-level, and aggregated community-level data at the spatial scale and year-to-minute level at the temporal scale. The datasets were consolidated from a comprehensive exploration of sources, including governments, research institutes, and online energy dashboards. The results of this effort include the aggregation of 60 datasets pertaining to building energy ontologies, building energy models, building energy and water data, electric vehicle data, weather data, building information data, text-mining-based research data, image data of buildings, fault detection diagnosis data and occupant data. A crowdsourcing mechanism in the platform allows users to submit datasets they suggest for inclusion by filling out an online form. This directory can fuel research and applications on building energy efficiency, which is an essential step toward addressing the world’s energy and environmental challenges.

Multi-objective scheduling of a steelmaking plant integrated with renewable energy sources and energy storage systems: Balancing costs, emissions and make-span

As an energy-intensive industry, the steel industry grapples with increasing energy costs and decarbonisation pressures. Therefore, multi-objective optimisation is widely applied in the production scheduling of the steelmaking plant. However, the optimal solution prioritising energy savings and emission reductions may lead to impractical or less economically efficient solutions, since the processing time requirement (PTR) of steel production orders in real-world production is neglected. This study fills the research gap by discussing the impact of PTR on the make-span of the steelmaking process and incorporating it into the optimisation model. Considering the variability of PTR, the solving of the multi-objective scheduling problem is transformed into the selection from Pareto solutions with different make-spans. To better leverage the temporal flexibility of the steelmaking process, a what-if-analysis-based strategy coupled with the Normal Boundary Intersection method is proposed to generate a series of evenly distributed Pareto solutions. The energy storage system is integrated to improve the time granularity of the steelmaking plant's flexibility. Our case studies demonstrate that the electricity and emission costs are reduced by 68.5%, indirect emissions are reduced by 83.5%, and the on-site renewable energy self-consumption rate increases by 12.1%. The effectiveness of the proposed method implies that it is of great relevance to the development of a cleaner steel industry in the future.

Network traffic characteristics of hyperscale data centers in the era of cloud applications

We present the network architecture of Alibaba Cloud DCs and investigate their traffic characteristics based on statistical data and captured traces. The statistical coarse-grained data are in the granularity of one minute, while the captured traces are fine-grained data that are in the granularity of one packet. We study the traffic features from the perspective of a macroscopic view, network performance, and microscopic view. The results report that the average utilization ratio of spine switches is stable when the observation time period reaches one day and the intra-ToR traffic ratio is in the range of 2%–10%. By mapping the folded-Clos topology to a tree topology and considering logical switching planes, we obtain the traffic matrix among pods from the average port utilization ratio. As we further investigate the perspective of network performance and the microscopic view, we find that there is no cell loss happening as the normalized queue speedQsis lower than 0.4. The normalized queue speedQsis defined as the total bytes of a queue sent in 1 s divided by 100 Gb, which reflects the packet sending speed of the queue. The observed maximum buffer size for one port conforms with the calculated maximum buffer occupation of 2.8 MB. By analyzing the captured traces, we find that the packet length is subject to a trimodal distribution. Under a time granularity of 10 ms, the instant bandwidth of one ToR port could reach 96 Gb/s at an average load of around 0.2 under a maximum link bandwidth of 100 Gb/s.

Bayesian Hyperparameter Optimization of stacked Bidirectional Long Short-Term Memory neural network for the State of Charge estimation

The battery industry has recently grown as a result of electric power adoption in many applications showing less reliance on fossil fuels. Accurate estimation of the State of Charge (SOC) is vital for optimal Battery Management System (BMS) operation to maintain the use of the battery within the safety limits. Data-driven methods require a careful hyperparameter selection as their results strongly depend on their choice. Heuristic methods involve manual tuning or exhaustive ways, which can be computationally demanding. Here a Bayesian Hyperparameter Optimization (BHO) with a Gaussian process is proposed, exploiting the benefits of the probabilistic approach to hyperparameter tuning to reduce the computational effort. For the accurate prediction of SOC, the BHO technique is implemented on a stacked Bidirectional Long Short-Term Memory (BiLSTM) neural network, incorporating a novel set of hyperparameters. The presented method is validated using two well-documented public datasets. Results are compared using datasets with different time granularity, revealing accuracy and computational differences while exhibiting superior yield capabilities compared to other state-of-the-art methods. To assess the computational effort associated with data size and model creation, a time granularity analysis was performed. The analysis demonstrated that decreasing the timestep resulted in reduced network creation time while maintaining similar error levels. Moreover, in order to ensure the comparability of the method, the metric of Floating-point Operations (FLOPs) was integrated to quantify the network’s computational volume. Finally, a sensitivity analysis performed on the time-series window span showed accuracy and computational load differences highlighting the need for caution and careful consideration.

Snapshot for Power Grids IoT: Adaptive Measurement for Resilience Intelligent Internet of Things

With the wide application of Internet of Things (IoT) devices in the power grids, the sophisticated feedback on their operating status is of great significance for improving efficiency and reducing accidents. For exquisite management of the resilient intelligent IoT with flexible increase and decrease of devices and heterogeneous operating systems, this paper proposes an adaptive measurement method ’MRAM’, which can snapshot the multidimensional resource view (MRV) of all devices in the jurisdiction. Extensible gateway platform based on CPU, FPGA and cloud computing is applied in MRAM, which liberates the local resources of monitored IoT devices. MRAM improves the LSTM algorithm called ELSTM. ELSTM can accommodate the current IoT devices’ state for detecting the mutation of MRV. The newly collected resources determined by ELSTM whether MRAM enters an abnormal state to drive the adaptive measurement state machine. According to the state machine which endeavors that the MRV is updated timely, MRAM adjusts the measurement granularity in real time. Simulations and experiments have tested the convergence time and occupied bandwidth of MRAM deployed in power grids. These evaluations confirmed MRAM’s practicality and robustness, as well as the MRV is genuine management data for the upper-layer power grids applications. A real environment is built to test the performance of this method as well. MRAM has high measurement accuracy and the precision of mutation detection is 98.41%. It converges the update MRV of second level under the condition of IoT devices and the cloud’s low consumption of memory and CPU utilization.

Modelling Temporality in Person- and Variable-Centred Approaches

Learning analytics needs to pay more attention to the temporal aspect of learning processes, especially in self-regulated learning (SRL) research. In doing so, learning analytics models should incorporate both the duration and frequency of learning activities, the passage of time, and the temporal order of learning activities. However, where this exhortation is widely supported, there is less agreement on its consequences. Does paying tribute to temporal aspects of learning processes necessarily imply that event-based models are to replace variable-based models, and analytic discovery methods substitute traditional statistical methods? We do not necessarily require such a paradigm shift to give temporal aspects their position. First, temporal aspects can be integrated into variable-based models that apply statistical methods by carefully choosing appropriate time windows and granularity levels. Second, in addressing temporality in learning analytic models that describe authentic learning settings, heterogeneity is of crucial importance in both variable- and event-based models. Variable-based person-centred modelling, where a heterogeneous sample is split into homogeneous subsamples, is suggested as a solution. Our conjecture is illustrated by an application of dispositional learning analytics, describing authentic learning processes over an eight-week full module of 2,360 students.

Mul-DesLSTM: An integrative multi-time granularity deep learning prediction method for urban rail transit short-term passenger flow

It is critical for the management and control of urban rail transit (URT) to be able to predict passenger flow accurately and in real time. Considering that the high-resolution data aggregated by the automatic fare collection (AFC) system is wasted, this paper analyzes the problem of applying a multi-time granularity passenger flow data fusion forecasting process. First, we examine the challenge of constructing a dataset of passenger flow data with different time granularities. Thus, an algorithm is proposed for selecting passenger flow datasets with multi-time granularity. Furthermore, a multi-time granularity dense residual network (Mul-DesLSTM) with a dense residual structure and LSTM (long short-term memory) as the predictor is constructed, inspired by a residual network. Using Mul-DesLSTM, finer-grained passenger flow features can be fused layer by layer while maintaining the accuracy of traditional single-granularity passenger flow predictions. Lastly, Mul-DesLSTM is applied to the URT system of Shanghai, China, and compared with baselines. As a result, the proposed Mul-DesLSTM outperforms the baselines with LSTM as a predictor and state-of-the-art model. When the predicted time granularity is 30 min, compared to the single-time granularity LSTM network, the mean absolute error, root mean square error, and symmetric mean absolute percentage error can be reduced by 51%, 63%, and 15%, respectively. The results can serve as a reference and basis for the operation and management of URT systems.

Timelines of adverse event journeys of LVAD patients.

The INTERMACS Events data set contains an expansive collection of temporal evidence of the course of adverse events (AEs) of >15 000 patients that have received a left ventricular assist device (LVAD). The chronology of AEs may contain insightful information of the "AE journeys" of LVAD patients. The purpose of this study is to investigate the timelines of AEs within the INTERMACS database. Descriptive statistics were applied to 86 912 recorded AEs of 15 820 patients with a continuous flow-LVAD between 2008 to 2016, extracted from INTERMACS registry. The characteristics of the timelines of AE journeys were investigated by posing six descriptive research questions. The analysis revealed several time-related characteristics and patterns of the AE journey after LVAD including the most common time of occurrences of AEs after surgery, duration of AEs journeys, the time of first and last AEs, and the time gaps between AEs. The INTERMACS Event dataset is a valuable resource for research about the timeline of AE journeys of patients who received an LVAD. It is necessary for future studies to first explore and consider the time-related characteristics of the data set such as diversity and sparsity to effectively choose an appropriate scope of time and time granularity and to acknowledge potential challenges.

Semi-supervised calibration of noisy event risk (SCANER) with electronic health records

ObjectiveElectronic health records (EHR), containing detailed longitudinal clinical information on a large number of patients and covering broad patient populations, open opportunities for comprehensive predictive modeling of disease progression and treatment response. However, since EHRs were originally constructed for administrative purposes not for research, in the EHR-linked studies, it is often not feasible to capture reliable information for analytical variables, especially in the survival setting, when both accurate event status and event times are needed for model building. For example, progression-free survival (PFS), a commonly used survival outcome for cancer patients, often involves complex information embedded in free-text clinical notes and cannot be extracted reliably. Proxies of PFS time such as time to the first mention of progression in the notes are at best good approximations to the true event time. This leads to difficulty in efficiently estimating event rates for an EHR patient cohort. Estimating survival rates based on error-prone outcome definitions can lead to biased results and hamper the power in the downstream analysis. On the other hand, extracting accurate event time information via manual annotation is time and resource intensive. The objective of this study is to develop a calibrated survival rate estimator using noisy outcomes from EHR data. Materials and MethodsIn this paper, we propose a two-stage semi-supervised calibration of noisy event rate (SCANER) estimator that can effectively overcome censoring induced dependency and attains more robust performance (i.e., not sensitive to misspecification of the imputation model) by fully utilizing both a small-labeled set of gold-standard survival outcomes annotated via manual chart review and a set of proxy features automatically captured via EHR in the unlabeled set. We validate the SCANER estimator by estimating the PFS rates for a virtual cohort of lung cancer patients from one large tertiary care center and the ICU-free survival rates for COVID patients from two large tertiary care centers. ResultsIn terms of survival rate estimates, the SCANER had very similar point estimates compared to the complete-case Kaplan Meier estimator. On the other hand, other benchmark methods for comparison, which fail to account for the induced dependency between event time and the censoring time conditioning on surrogate outcomes, produced biased results across all three case studies. In terms of standard errors, the SCANER estimator was more efficient than the KM estimator, with up to 50% efficiency gain. ConclusionThe SCANER estimator achieves more efficient, robust, and accurate survival rate estimates compared to existing approaches. This promising new approach can also improve the resolution (i.e., granularity of event time) by using labels conditioning on multiple surrogates, particularly among less common or poorly coded conditions.

Two-Timescale Learning-Based Task Offloading for Remote IoT in Integrated Satellite–Terrestrial Networks

In this paper, we propose an integrated satellite-terrestrial network (ISTN) architecture to support delay-sensitive task offloading for remote Internet of Things (IoT), in which satellite networks serve as a complement to terrestrial networks by providing additional communication resources, backhaul capacities, and seamless coverage. Under this architecture, we investigate how to jointly make offloading link selection and bandwidth allocation decisions for BSs and IoT users. Considering the differentiated decision-making time granularities, we formulate a two-timescale stochastic optimization problem to minimize the overall task offloading delay. To accommodate the two-timescale network dynamics and characterize state-action relations, we establish a hierarchical Markov decision process (H-MDP) framework with two separate agents tackling two-timescale network management decisions, and two evolved MDP-based subproblems are formulated accordingly. To efficiently solve the subproblems, we further develop a hybrid proximal policy optimization (H-PPO)-based algorithm. Specifically, a hybrid actor-critic architecture is designed to deal with the mixed discrete and continuous actions. In addition, an action mask layer and an action shaping function are designed to sample feasible task offloading decisions from the time-variant action set. Extensive simulation results have validated the superiority of the proposed ISTN architecture and the H-PPO-based algorithm, especially in scenarios with scarce spectrum resources and heavy traffic loads.

Hybrid prediction model for cold load in large public buildings based on mean residual feedback and improved SVR

In order to control the energy efficiency of air conditioning systems and design energy management strategies, accurate and effective prediction of building cooling demands is a crucial step. We propose a time-granular residual feedback-based and improved sparrow algorithm for optimizing support vector regression models (TGRF-ISSA-SVR). To begin, select the input features using Random forest. Second, a method (TGRF) is proposed to calculate and feedback the mean residual with time granularities of hour, day, and month. Then, three strategies are used to improve the Sparrow Algorithm (ISSA) and apply it to SVR parameter optimization. Finally, the measured data of a large commercial building in Xi'an are used for experimental verification. The RMSE of TGRF-ISSA-SVR is 4.33 and the MAPE is 0.66, indicating that the model has low error. The RMSE and MAPE of TGRF-ISSA-SVR are 75% and 74% lower than those of TGRF-SVR. The RMSE and MAPE of TGRF-SVR are 37% and 38% lower than those of SVR, demonstrating that the improved model has accuracy improvement. In addition, compared with models such as LSTM and GRNN, the TGRF-ISSA-SVR model has higher prediction accuracy and shorter prediction time, so TGRF-ISSA-SVR can provide effective support for load prediction of large public buildings.

Neural computing for grey Richards differential equation to forecast traffic parameters with various time granularity

The existing traffic parameter prediction methods generally adopt a single prediction model, but the fusion of different theories and methods can complement each other and improve the prediction performance of the model. Starting from the statistical distribution characteristics of traffic flow, this work introduces the Richards equation to conduct grey modeling, which is used to simulate the development trend of traffic parameters; and then fuses the abilities of error feedback adjustment and complex nonlinear fitting of neural network to estimate the parameters of the grey model and forecast the volatility of traffic flow respectively. At the same time, a dynamic prediction framework of real-time data update is built for the new model, and finally establish the dynamic grey Richards neural network model (DGR-NN).Apply the new model to different traffic parameters (Speed; Volume; Jam mileage) and data resolutions (5 min; 15 min; 1 h), the modeling effect of DGR-NN is significantly improved compared to the grey Richards model (GRM), and the training and testing errors of the model in the three forecast scenarios are reduced to varying degrees, where the testing MAPE, RMSE and STD are reduced by 1.80% ∼ 10.87%, 2.55% ∼ 7.26% and 8.08–25.56% respectively. Furthermore, the results of the new model were verified and analyzed with the other five comparison models, among which the boxplot of APE shows that the error distribution of DGR-NN prediction data is concentrated and the value level is relatively low. It can be seen that DGR-NN can accurately and stably forecast traffic parameters of different time granularities.

Load-forecasting method for IES based on LSTM and dynamic similar days with multi-features

To fully exploit the rich characteristic variation laws of an integrated energy system (IES) and further improve the short-term load-forecasting accuracy, a load-forecasting method is proposed for an IES based on LSTM and dynamic similar days with multi-features. Feature expansion was performed to construct a comprehensive load day covering the load and meteorological information with coarse and fine time granularity, far and near time periods. The Gaussian mixture model (GMM) was used to divide the scene of the comprehensive load day, and gray correlation analysis was used to match the scene with the coarse time granularity characteristics of the day to be forecasted. Five typical days with the highest correlation with the day to be predicted in the scene were selected to construct a “dynamic similar day” by weighting. The key features of adjacent days and dynamic similar days were used to forecast multi-loads with fine time granularity using LSTM. Comparing the static features as input and the selection method of similar days based on non-extended single features, the effectiveness of the proposed prediction method was verified.

Ev-Conv: Fast CNN Inference on Event Camera Inputs for High-Speed Robot Perception

Event cameras capture visual information with a high temporal resolution and a wide dynamic range. This enables capturing visual information at fine time granularities (e.g., microseconds) in rapidly changing environments. This makes event cameras highly useful for high-speed robotics tasks involving rapid motion, such as high-speed perception, object tracking, and control. However, convolutional neural network inference on event camera streams cannot currently perform real-time inference at the high speeds at which event cameras operate—current CNN inference times are typically closer in order of magnitude to the frame rates of regular frame-based cameras. Real-time inference at event camera rates is necessary to fully leverage the high frequency and high temporal resolution that event cameras offer. This paper presents Ev-Conv, a new approach to enable fast inference on CNNs for inputs from event cameras. We observe that consecutive inputs to the CNN from an event camera have only <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">small differences</i> between them. Thus, we propose to perform inference on the difference between consecutive input tensors, or the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">increment</i> . This enables a significant reduction in the number of floating-point operations required (and thus the inference latency) because <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">increments</i> are very sparse. We design Ev-Conv to leverage the irregular sparsity in increments from event cameras and to retain the sparsity of these increments across all layers of the network. We demonstrate a reduction in the number of floating operations required in the forward pass by up to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$98\%$</tex-math></inline-formula> . We also demonstrate a speedup of up to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$1.6\times$</tex-math></inline-formula> for inference using CNNs for tasks such as depth estimation, object recognition, and optical flow estimation, with almost no loss in accuracy.