Financial Credit Risk Management
Published:
Blog posts
2024
Fully Convolutional Networks
Published:
Fully Convolutional Networks for Semantic Segmentation
GitHub
- 去除全连接层： 传统的卷积神经网络（CNN）通常包含卷积层和全连接层。在这篇论文中，作者认识到全连接层会导致输出固定大小的特征向量，从而限制了CNN在图像语义分割任务中的应用。 因此，作者去除了CNN中的全连接层，将其替换为全卷积层，使得网络能够接受任意大小的输入图像，并输出相同尺寸的像素级别的分类结果。
Convolutional Neural Networks
Published:
什么是卷积神经网络（CNN）？它在计算机视觉中的应用是什么？
Basic pandas
Published:
Python, data science, pandas
Basic numpy
Published:
Python, data science, numpy
Happy 31th birthday
Published:
今天不谈机器学习，不再整理一些记不住的知识，而是在三十又一的当下，记录一下最近的思考。
Clustering alrogithms
Published:
gradient descent, stochastic gradient descent, batch gradient descent
Activation Functions
Published:
激活函数（Activation Function），负责将神经元的输入映射到输出端，激活函数将神经网络中将输入信号的总和转换为输出信号。激活函数大多是非线性函数，才能将多层感知机的输出转换为非线性，使得神经网络可以任意逼近任何非线性函数，进而可以应用到众多的非线性模型中。
Loss Functions
Published:
Regression Loss Functions
Classification Algorithms
Published:
Classification Algorithms
Time-series Forecasting
Published:
ARIMA (Autoregressive Integrated Moving Average)
It is a popular statistical analysis model used for forecasting time series data. ARIMA models are especially well-suited for short to medium-term forecasting models that have data with trends, seasonality, or cyclic patterns. The model aims to describe the autocorrelations in the data.
Clustering alrogithms
Published:
The importance of clustering are of
Metrics to Evaluate Predictive Models
Published:
Metrics used to evaluate predictive modeling, highly used in regression, time-series forecasting cases.
Donut: an OCR-free Document Understanding Transformer
Published:
My notes on this paper: OCR-free Document Understanding Transformer (github repo).
Object Detection Algorithms
Published:
A summary of object detection algorithms