• Trust learning — traits


    Traits: defining common behaviors — Characteristics Features tell the rust compiler what a particular type has and can be shared with other types. We can use features to define sharing behavior in an abstract way. We can use feature ranges to specify that generics can be any type with a specific behavior. Note: features are […]

  • Deep residual shrinkage network


    In this paper, a new deep learning algorithm deep residual shrinkage network is discussed. 1. The basis of deep residual shrinkage network It can be seen from the name that the deep residual shrinkage network is an improved method of the deep residual network. Its feature is “shrinkage”, which refers to soft thresholding, which is […]

  • Interpretation of didi ETA paper: WDR model


    introduction paper:Learning to Estimate the Travel Time Impressions after reading: the overall standard, the combination of WD model and LSTM, to solve practical business problems. Note: ETA is the abbreviation of estimate travel time, that is, the estimated time of arrival. The problem is to estimate the time from point a to point B at […]

  • Overview of map search system


    Overview of map search system Image search by image refers to searching images with similar content according to image content. Building a map search system needs to solve two key problems: first, extracting image features; second, feature data search engine, that is, building feature data into database and providing the function of similarity search. Image […]

  • Group normalization + pytorch code


    The article is transferred from official account (machine learning alchemy), and it pays attention to “alchemy” to get massive free learning materials. In general, GN is the improvement of BN and the balance of in and LN. Advantages of 1 bn Here is a brief introduction of BN. In the previous article, we have introduced […]

  • Attempt of multi task learning in real time delivery


    introduction Now most machine learning tasks are single task learning. For complex problems, it can also be decomposed into simple and independent subproblems to solve separately, and then the results are combined to get the results of the initial complex problems. This seems reasonable, but in fact it is not correct, because many problems in […]

  • Convolutional neural network CNN learning 1


    Convolutional neural network CNN learning 1   Ten years of sharpening a sword, the frost blade has never been tried.   Introduction:Convolutional neural network CNN learning. CNN Chinese video learning link:Convolutional neural network working principle Video – Chinese version CNN English learning link:Convolutional neural network working principle video 1、 Definition Convolutional neural networks is a […]

  • Borderdet: greatly improve the detection accuracy through boundary features, plug and play, and the speed is not slow | ECCV 2020 oral


    Boundary is very important for location problem. Borderalign, the core idea of borderdet, is ingenious and effective. It not only integrates boundary features into target location prediction, but also can be easily integrated into various target detection algorithms, which brings great performance improvement. In the open source implementation, the efficient CUDA implementation of borderalign will […]

  • [IOS / swift] Bluetooth connection


    I have done Bluetooth connection function before, but either I change and add new functions directly on the basis of others, or I just follow others’ articles step by step. After I finish writing, I still have no idea, just can use it. This time, we take the opportunity of designing the ymodem upgrade tool […]

  • Jitter recommendation algorithm principle tiktok


    Jitter recommendation algorithm principle tiktokThis sharing will mainly introduce the overview of today’s headline recommendation system and the principles of content analysis, user tags, evaluation analysis, content security, etc. 1、 System overview If the recommendation system is described in a formal way, it is actually a function fitting a user’s satisfaction with the content. This […]

  • Fast, developed by gojek and Google cloud, joined lf AI & data as an incubation project


    Author: Christina Harter The LF AI & data foundation, which is building an ecosystem to support open source innovation in artificial intelligence (AI), machine learning (ML), deep learning (DL) and data open source projects, today announced fast as its latest incubation project.Feast(Feature StMore (feature store) is an open source feature store for machine learning. Today, […]

  • Sparkml predicts PV


    background The company needs to forecast the daily traffic for a period of time in the future according to the daily traffic data of the website in the past period, so that the relevant operation and operation and maintenance can be alerted in advance before the peak traffic. This is a typical “time series prediction […]