normal OLS regression

MLE assumption 问题 MLE的成立不建立在任何假设上。有时MLE最后的形式不是一个closed form,不能确切的求出这个estimator。所以这个时候,我们可以asymptotically来approximate 这个值。当n 趋于无穷的时候,MLE会follow一个normal distribution: n(θ^−θ)∼N(0,I−1)\sqrt n(\hat\theta-\theta)\sim N(0, ... 阅读详情


A3. Advertising Content Audit – Individual Report [CLO1, CLO2, CLO3, CLO4]
Due: Sunday (November 17th) 11.55 PM / Weighting: 40% / Length: Max 2,000 words (+/- 10%)
Description 
This assessment provides you with the opportunity to undertake an advertising content audit to evaluate whether a company’s current advertising strategy is proper and to provide recommendations for further improvements. 
You are asked to evaluate the consistency of beauty and cosmetics company mamaearth’s Instagram posts (https://www.instagram.com/mamaearth.in/) with its overall positioning objectives. In addition, by analyzing Instagram content and social media engagement data (e.g., the number of likes, comments), you can evaluate whether the theme related tothe company's positioning is more successful than other themes in generating engagement. 
Details
1) The focal company is mamaearth.in which is an Indian beauty, cosmetics and personal care brand. Identify the company’s positioning or value proposition statement (or suitable summary) and provide it (with relevant citation) onthe title page of your report. (Note: there are many companies who use the name “mamaearth” worldwide. We are investigating the company with the instagram account as given above)
 
Follow the below steps to complete your task:
 
2) Consistent Advertising with Positioning: We want to identify relevant themes across posted images. For the purpose of this assignment, we will assume the relevant main themes are reasonably consistent (but not identical) with the main themes introduced in the Bodyshop case study (Lecture Week 8); Product, People, Baby, Other, About Positioning. 
 
3) You have been provided with the “mamaearth” instagramimages, PB metadata for the images and data extracted from Google Vision (Ocr, label, multiple objects). 
Main Folder – Assignment 3_T3_24   Link: 
Images sub-folder – 1619 mamaearth Instagram post images;  
Extracted Vision sub-folder – label_data.xlsx, multiple_objects_data.xlsx, ocr_data.xlsx
PB metadata - mamaearth_PB_T3_24_ Ass3(1619).xlsx
 
Initially, you will need to;
(i) Construct OpenCV data using the images (2.3 Image processing using OpenCV)
(ii) Process the captions/description in the PB metadata file, to obtain Valence_overall and length_descriptionand add them to PB metadata file. (4.2 Text Analysis Variables – will help). 
 
 
 
 
Since the relevant themes are somewhat consistent with the Bodyshop case, you have some base for constructing a dictionary for the main themes. You can add to this base or modify, by sorting detected features from Google Vision according to their usage frequency and adding any items which are specific to mamaearth (possibly deleting items which are specific to the Bodyshop), if required. (This may be more relevant for the “About Positioning” theme since the positioning of the Bodyshop may vary from the positioning of mamaearth) Using the constructed dictionary, determine which themes are present in the posted mamaearth images. 
 
(Do some preliminary googlevision processing with the multiple_objects, label and ocr googlevision data – 5.2 Construct Variables from Googlevision will help)
 
From the preliminary googlevision processing, construct your data dictionary for the 5 key themes (use the bodyshopdictionary as a base) which you will then use in 8.1 Main Category_Variables_googlevision notebook.
 
4) Report at least two posts (screenshots of images and text caption) about each theme. Also, state how many posts (frequency and percentage) are related to each theme. Based on this usage frequency, discuss what you suggest the company’s advertising content strategy is. In addition, evaluate whether you feel the frequency of the positioning-related theme is proper. Why? Or why not? 
 
 
5) Advertising Effectiveness: (Evaluate whether the positioning-related theme is more successful compared to other popular themes in social media engagement by doing the following tasks): 
 
Construct the regression data set which contains OpenCV, PB metadata and theme categories as dummy variables (also include relevant control variables: (1) text length in the caption, (2) text sentiment in the caption, (3) OCR text length within an image, (4) OCR text sentiment within an image, and (5) Posting time dummies: Year, Month-of-Year (January, …, December), Day-of-Week (Monday, ..., Sunday), Time-of-day (Morning, Afternoon, Evening, Night). 
 
Report summary statistics (count/frequency, mean, median, minimum, maximum) in a Table for each of the X and Y (like count, comment count) variables. 
 
For Y equal to “like count” use normal OLS regression (considering multicollinearity) on either like count or Ln (like count+ 1) or both.
Report the result of your model (s) in a Table (columns: X variables, coefficients, p-value, and VIF). Interpret the results focusing on the significance and relative importance of the separate themes in impacting viewer engagement.
 
For Y equal to “comment count” use normal OLS, Poisson and Negative Binomial regression.
Explain your choice of candidate regression models. For comment count report the regression result of the best model in a Table (columns: X variables, coefficients, p-value, and VIF). Interpret the results focusing on the significance and relative importance of the separate themes in impacting viewer engagement.
 
Effective Ad Content Exploration: Now, you are required to break down the positioning-related theme into several (max 5) sub-categories (See the examples in class materials). For example, there may be sub-themes based on “Organic” or “Sustainable” which represent how the organization wants consumers to view the brand. You can refer to the mission statement and your overview of the organization’s positioning to provide potential sub-categories (sub-themes). Provide your rationale for this sub-categorization (ensure it is specific to mamaearthand its positioning). 
 
Report at least two posts, the number of posts, and summary statistics for each sub-category. 
 
For Y equal to “like count” use normal OLS regression (considering multicollinearity) on either like count or Ln (like count+ 1) or both.
 
Report the result of your model (s) in a Table (columns: X variables, coefficients, p-value, and VIF). Interpret the results focusing on the significance and relative importance of the separate positioning sub-themes in impacting viewer engagement.
 
For Y equal to “comment count” use normal OLS, Poisson and Negative Binomial regression.
Explain your choice of candidate regression models. For comment count report the regression result of the best model in a Table (columns: X variables, coefficients, p-value, and VIF). Interpret the results focusing on the significance and relative importance of the separate positioning sub-themes in impacting viewer engagement.
 
Construct your data dictionary for the 3 to 5 sub-positioning themes (use the Positioning dictionary as a base and sub-divide into 3 to 5 sub-categories) which you will then use in 8.2 Sub Category_Variables_googlevisionnotebook.
 
Merge all the data (OpenCV, PB metadata, Main category data (from 8.1), Sub-category data (from 8.2) into one regression ready file (8.3 Merge_All_Variables_Regression_data)
 
Use the regression file which will come from 8.3 to do the various regressions.  Use 8.4 Regression_models with main & sub-category variables (T3_24) to conduct the regressions. (You may need to run this notebook several times (modifying the code) to obtain the required regressions for each part)
 
 
 
6) Conclusion: Provide conclusions on which themes the company is focussed. Is their advertising content consistent with the company’s positioning? Which positioning (sub) themes are successful? Considering all of the above, recommend what the company needs to keep doing or improvements it needs to make to enhance engagement and consistency with positioning. 
In completing this task, apply appropriate data analytics and consider the concepts introduced in class. Your report should not exceed the word limit, excluding the title page, relevant images, tables or charts.     
Title page (1 page) includes (1) Company & Positioning statement, (2) Word count, (3) An executive summary (One paragraph) of your report, (4) Course name, tutorial session and group, tutor’s name, (5) Your first and last name & zID
Reference: (If any) Cite academic papers, newspaper articles, blogs, or industry reports using Endnote. Use APA (American Psychological Association) style in-text citations and a reference list at the end. https://student.unsw.edu.au/apa
Format: Use word file (.doc), 12pt, 1.5 lines spacing, at least 2.5cm margins on all sides.
 
Submission instructions
Submit your report to Turnitin via Moodle to the Reportsubmissions folder.
1) .doc contains your report. File name: Tutorial_Group_Firstand Last Name_A3.doc” (e.g., W12_1_Con Korkofingas_zXXXX_A3.doc)
 
Submit other supporting files (data, image, paper and code) to the Other Files submission folder. 
2) .xlsx file contains the dataset on which you run yourregressions.
3) .ipynb contains all relevant code to get the results in your report. Make a zip file by combining all colab files. 
4) .xlsx files – other relevant excel files (dictionary, summary statistics etc, relevant base files)
 
● For each missing file among the above (2) to (4), -1 mark  
 
Marking Criteria
Your assignment will be marked based on the following marking criteria:
1. Analysis: Quality of advertising image data analytics 
2. Evaluation and Recommendations: Quality of evaluation and recommendations
3. Written Presentation: Quality of written report
For further information, see the below marking rubric. 
Marking Rubric for Assessment 3: Advertising Content Audit – Individual Report
Criteria     %     Fail      Pass      Credit      Distinction      High Distinction  
Analysis 
Quality of advertising image data analytics    40%     Analysis of advertising image data does not meet the required standard. 
 
 
     Sufficient identification of popular advertising themes. Mostly appropriate analysis of advertising data but may not include all of the criteria, i.e. categorisation, variable construction, and/or statistical testing. 
     Appropriate identification of popular advertising themes, including one related to the company’s positioning statement; good and mostly appropriate analysis, categorisation, variable construction, and statistical testing of advertising data.    Excellent and accurate identification of popular advertising themes, including one related to the company’s positioning statement; effective and proper analysis, categorisation, variable construction and statistical testing of advertising data.    Excellent and highly accurate identification of popular advertising themes, including one related to the company’s positioning statement; highly effective and proper analysis, categorisation, variable construction and statistical testing of advertising data. 
Evaluation & Recommendations 
Quality of evaluation and recommendations    40%    Evaluation and recommendation do not meet the required standard.     Social media advertising data is sufficiently evaluated with some conclusions and an opinion provided. Recommendations are provided.     Social media advertising data is accurately evaluated to draw conclusions, provide an opinion, and determine actions. Recommendations are mostly appropriate and justified with some evidence from data and course concepts.    Social media advertising data is accurately evaluated to draw conclusions, provide an opinion, and determine actions. Recommendations are appropriate and justified with specific evidence from data, course concepts and scholarly papers.    Social media advertising data is accurately and meaningfully evaluated to draw conclusions, provide an opinion, and determine actions. Recommendations are highly appropriate, actionable, justified with specific evidence from data, course concepts, scholarly papers, and industry articles and offer new insight.
Written Presentation
Quality of written report     20%    Report lacks clear structure. Written English is below the required standard.
 
​    Report provides a mostlyappropriate structure with distinguishable paragraphs. Written English is appropriate to the task but has spelling, referencing and/or grammatical errors    Report is clearly structured with good transitions and paragraphs. Good use of written English, which is appropriate to the task and has few spelling, referencing and/or grammatical errors. Report mostly adhered to the prescribed word count and conventions.    Report is clearly structured with excellent transitions. Above standard use of written English language, which is professional and appropriate to the task with minimal spelling, referencing and/or grammatical errors. Report adheres to the prescribed word count and conventions.    Report is clearly and logically structured with excellent transitions and paragraphs. Excellent use of written English language, which is professional and appropriate to the task and has minimal spelling, referencing, and/or grammatical errors. Report adheres to the prescribed word count and conventions.
 

Statsmodels深度解析:从统计推断到时间序列预测的Python实战指南 统计推断是数据分析的核心基础,它关注变量间关系的显著性、效应大小及模型假设检验,而非仅仅追求预测精度。其原理基于概率论与数理统计,通过假设检验、置信区间估计等方法,为从样本数据推断总体特征提供严谨的数学框架。这一技术价值在于能够提供可解释、可复现的结论,是科学研究、商业决策和金融计量等领域做出可靠判断的基石。在应用场景上,统计推断广泛应用于线性回归、广义线性模型(如逻辑回归)以及时间序列分析(如ARIMA模型)中,用于探究因果关系、评估政策效果或进行经济预测。本文聚焦于Python中的Statsmodels 阅读详情

相关推荐

Stata入门实战:从数据导入到回归分析的全流程操作指南

在实证研究与数据分析领域,掌握高效的数据处理与统计软件是开展研究的基础。Stata作为该领域的核心工具,其基于命令行的操作模式,将复杂的数据清洗、变量生成与计量分析封装为简洁的指令,极大地提升了研究效率与可复现性。理解其以‘命令、变量、观测值’为核心的数据结构和工作原理,是有效使用该软件的关键。这种操作逻辑的价值在于,它允许研究者通过编写脚本(Do文件)完整记录分析过程,确保从数据导入、异常值处理到模型估计的每一步都可追溯、可重复。无论是处理微观调查数据还是宏观面板数据,这一流程都是构建稳健实证分析的基石。

weixin_34198762的博客 632

Logistic Regression实战指南:业务可解释预测模型落地全流程

Logistic Regression是一种基础但强大的广义线性模型,其核心本质是建模事件发生的条件概率,而非简单分类。它通过logit链接函数实现[0,1]区间内的概率校准,具备天然的可解释性与审计友好性,特别适用于金融风控、用户流失预警、营销响应预测等需向业务方清晰归因的场景。相比XGBoost或深度学习,它在资源受限、系统兼容性差、监管要求高(如GDPR/银保监规则)的环境中更具工程落地优势。本文聚焦真实业务约束下的建模实践,涵盖数据清洗陷阱识别、业务驱动的特征构造(如‘距今天数’分段)、共线性诊断与

weixin_34279246的博客 527

线性回归实战:从数据生成到残差诊断的工程化建模

线性回归是机器学习中最基础且应用最广的统计建模方法,其核心在于通过最小二乘法拟合特征与目标变量间的线性关系,并依托正态性、同方差性、无多重共线性等经典假设保障推断可靠性。该模型的技术价值不仅体现在预测精度,更在于系数可解释性、部署轻量性与业务归因能力——这使其在金融风控、销售预测、运营优化等强合规、高审计要求场景中不可替代。典型应用场景包括门店销售额预测、客户生命周期价值(LTV)估算、设备故障率建模等。本文聚焦真实项目中的‘诊断驱动建模’范式,系统覆盖数据仿真、标准化陷阱、VIF动态阈值设定、Box-Co

turnRound 1084

StatsModels加权最小二乘法:3个实战技巧优化异方差数据回归

你有没有遇到过这样的情况?用普通最小二乘法(OLS)拟合数据时,模型看起来不错,但预测区间总是"不对劲"——在某些区域的预测过于自信,在另一些区域又过于保守。这很可能是因为你的数据存在**异方差性**,而StatsModels的加权最小二乘法(WLS)正是解决这个问题的利器。 ## 异方差:数据中隐藏的不平等 想象一下,你在分析不同规模公司的财务数据。小公司的营收波动可能只有几十万,而大公司的

gitblog_01195的博客 481

最小化均方误差(MSE)与频繁主义建模实战指南

最小化均方误差(MSE)是预测建模中最基础且广泛应用的优化目标,其核心在于通过平方误差量化预测偏差,兼顾对大错误的敏感性与数学可解性;在频繁主义框架下,MSE估计具有无偏性、有效性与一致性等严格统计保障,适用于需稳定复现与可验证结果的工程场景;该方法天然适配高斯噪声假设,并与最大似然估计一致,广泛用于线性回归、传感器校准、风险评估与实时决策系统;本文结合正规方程推导、偏差-方差分解、正则化调优及生产监控,系统呈现从理论原理到工业级落地的完整技术链条。

weixin_34323858的博客 393

柯西分布与重尾数据建模:从异常识别到稳健分析

重尾分布是真实世界数据的普遍特征,其尾部衰减缓慢、极端值频发,导致传统基于正态假设的统计方法(如t检验、OLS回归、3σ阈值)严重失真。柯西分布作为典型稳定重尾分布,无期望与方差,尾部按x⁻²衰减,天然适用于高频交易滑点、网络延迟峰值、金融风险等场景。理解其参数意义(中位数代替均值、MAD代替标准差)、掌握QQ图与Hill估计器诊断方法、采用分位数阈值与柯西回归等稳健技术,是构建抗干扰数据模型的关键路径。本文聚焦工程落地,提供可复现的Python分析流水线。

weixin_30591551的博客 392

StatsModels加权最小二乘法实战指南:异方差数据的高效处理方案

在现实数据分析中,异方差性就像数据分析中的"隐形杀手"——它不会让模型完全失效,但会悄悄降低估计效率,让统计推断变得不可靠。StatsModels作为Python统计建模的瑞士军刀,提供了强大的加权最小二乘法(WLS)工具来解决这一难题。本文将带你从实际问题出发,深入探索WLS在异方差场景下的实战应用。 ## 异方差:数据中的"不公平"现象 想象一下这样的场景:在金融数据中,大额交易的波动性往

gitblog_00830的博客 242

手写普通最小二乘法:从矩阵推导到残差诊断的线性回归实战

线性回归是监督学习的基石,其核心原理源于普通最小二乘法(OLS)——一种通过最小化残差平方和来求解最优参数的数学优化方法。该方法依托矩阵运算(如XᵀX求逆、正交投影)实现解析解,不仅提供可解释的系数估计,更构成模型可解释性与残差分析的理论基础。相比黑盒调用,手写OLS能暴露数据病态性(如多重共线性)、揭示R²局限性,并以残差图作为诊断‘X光机’,支撑工业级模型校准与调试。本文聚焦从正规方程推导、数值稳定实现到统计推断的完整链路,适用于夯实机器学习底层逻辑的工程师与数据科学学习者。

weixin_34062469的博客 409

StatsModels加权最小二乘法:解决异方差问题的核心技术解析

在现实世界的数据分析中,我们经常遇到一个棘手的问题:数据的波动性并不均匀。想象一下金融交易数据——小额交易的波动性通常较小,而大额交易的波动性却可能急剧增加。这种数据特征在统计学中被称为"异方差性",而加权最小二乘法(WLS)正是StatsModels为解决这一问题提供的强大工具。 ## 异方差性:回归分析中的隐形杀手 异方差性是线性回归模型中最常见的假设违反之一。当误差项的方差随自变量的变化

gitblog_00197的博客 418

线性回归中Type I错误的实战解析:多元场景下的p值陷阱

线性回归是数据分析与建模最基础的统计工具,其核心依赖于假设检验——尤其是对回归系数是否为零的显著性判断。然而,当模型包含多个自变量时,经典p值检验会因多重比较而系统性放大第一类错误(Type I Error)风险,即把纯随机噪声误判为真实效应。这种错误并非计算失误,而是统计推断框架固有的概率代价,在金融风控、销量预测、故障预警等工程实践中极易导致‘显著却无效’的决策误导。本文通过可控模拟实验,揭示变量数量、样本量与多重共线性如何协同扭曲p值解释,并引入Bonferroni与FDR等校正策略及效应量评估,帮助

duiker 395

避开统计陷阱:当你的数据彼此关联时,为什么不能用普通回归?SPSS GEE保姆级指南

本文深入探讨了当数据存在关联性时为何不能使用普通回归分析,并提供了SPSS中广义估算方程(GEE)的详细操作指南。通过实际案例和模拟数据,解释了非独立相关数据对统计结果的致命影响,帮助研究者避免常见统计陷阱,提升数据分析的准确性。

weixin_30667649的博客 345

用线性回归+虚拟变量精准定位时间序列趋势突变点

趋势突变检测是时间序列分析中的基础任务,其核心在于识别序列中长期趋势发生永久性偏移的关键时刻。传统方法依赖人工切片或滑动窗口,易受主观偏差与噪声干扰;而基于统计推断的回归建模方法,通过引入阶跃型虚拟变量(dummy variable)将突变点检测转化为系数显著性检验问题,天然具备全局信息利用、抗噪性强、可解释性高三大优势。该方法以OLS回归为底座,以z值为判据,无需深度学习框架,适用于分钟级至日级业务指标监控,在电商转化率、SaaS留存斜率、API成功率等趋势主导型场景中已验证实效。本文即围绕这一轻量但鲁棒

weixin_30315723的博客 551

Rust机器学习实战:回归与分类的生产级实现

机器学习模型部署面临Python生态的运行时不确定性、内存管理脆弱性和跨平台兼容性挑战。Rust凭借零成本抽象、编译期类型安全和静态链接能力,为回归与分类任务提供高确定性数值计算基础。其核心在于ndarray的内存布局优化与linfa的trait-based算法设计,兼顾SIMD向量化性能与FFI互操作性。在金融风控、边缘推理和实时特征工程等对延迟、资源与可靠性敏感的场景中,Rust可显著降低GC停顿、内存抖动与依赖冲突风险,成为Python实验后走向生产环境的关键技术跃迁路径。

a1311010193的博客 400

多变量线性回归:一因多果协同建模的核心原理与工程实践

多变量线性回归是一种处理多个相关因变量(Y)同时受同一组自变量(X)影响的统计建模方法,其核心在于建模因变量间的残差协方差结构(Σ),而非简单重复单变量回归。它基于联合最小二乘估计,通过矩阵形式 Y = XB + E 统一刻画多任务响应机制,在保持参数无偏性的同时显著提升估计效率与假设检验可靠性。该技术在工业过程控制、生物医学监测、金融风险联动等需同步评估强耦合指标的场景中具有不可替代的技术价值,是理解‘一因多果’系统行为的基础建模范式。

342

岭回归估计量的分布近似与置信区间构造详解

在回归分析中,岭回归通过L2惩罚解决多重共线性问题,以牺牲无偏性换取更低的估计方差。对于需要输出回归系数置信区间的工程场景,理解估计量的抽样分布至关重要。固定岭参数时,估计量可视为OLS的线性变换,服从正态分布;但实际中参数常由数据自适应选择,导致简单正态近似存在偏差。本文从均值、方差推导出发,结合Monte Carlo模拟验证近似效果,并对比偏差修正与Bootstrap等区间构造方法,帮助数据科学实践者合理评估估计量的不确定性与模型稳健性。

weixin_34006468的博客 383

逻辑回归实战:从概率校准到业务可解释的全流程工程指南

逻辑回归不仅是基础二分类算法,更是工业界广泛采用的概率校准引擎与可解释性建模范式。其核心在于logit变换对伯努利分布的自然适配,以及最大似然估计对分类边界的精准聚焦。相比线性回归+阈值的粗放处理,逻辑回归通过sigmoid约束输出为有效概率,并赋予系数明确的业务语义——如优势比(Odds Ratio)可直译为‘某特征每增加一单位,事件发生几率变化倍数’。在银行风控、医疗预警、推荐系统等高可信场景中,它支撑着评分卡构建、边缘部署与联邦学习等关键实践。本文聚焦生产级落地,覆盖缺失值鲁棒编码、目标编码平滑、L1

weixin_34344403的博客 379

回归分析实战:从数据探索到模型诊断的完整建模指南

回归分析是数据分析与预测建模的基石,其核心思想是通过最小二乘法寻找变量间的最佳数学关系。该方法旨在量化自变量与因变量之间的关联,从而将业务问题转化为可计算的模型。其技术价值在于提供了一套从数据中提取规律、进行科学预测与决策的标准化框架,广泛应用于销售预测、成本分析、市场研究等场景。在实践过程中,模型诊断与优化至关重要,需重点关注残差分析以检验模型假设,并利用方差膨胀因子(VIF)诊断多重共线性问题,确保模型稳健可靠。本文以Python代码为例,系统演示了从数据可视化、模型拟合、统计检验到残差分析与共线性诊断

anjichan4261的博客 513

我的AI学习 ai维护工具之langfuse

as_type: Optional[Literal[‘generation’]] = None, # 将记录定义为 Observation (LLM 调用)transform_to_string: Optional[Callable[[Iterable], str]] = None # 将输出转为 string。要用LangFuse的云服务,首先要在LangFuse官网上注册,注册完后将页面拉倒最后,点免费的SignUp,API文档:https://api.reference.langfuse.com/

liubaobin的博客 4096

金融数据分析实战:Pandas数据处理与回归策略全攻略

在Python数据分析领域,数据处理是核心基础。Pandas作为最流行的结构化数据工具,为金融统计与量化交易提供了高效的数据清洗、转换与聚合能力。本文系统梳理了从Pandas基础操作到回归建模、策略回测的完整学习路径,涵盖环境配置、核心语法、真实行情数据获取、描述性统计、OLS回归解读以及动量策略构建等关键环节。通过可复现的代码示例,帮助读者掌握避免前视偏差、处理缺失值、计算滚动波动率等工程实践,并深入理解Beta系数、夏普比率等指标的业务含义。无论你是零基础入门还是希望提升量化分析能力,都能从中获得从数据

weixin_30680385的博客 438

汽车底盘技术解析:从悬架调校到智能化,如何评价国产车底盘实力

汽车底盘作为决定车辆驾驶质感、安全性和耐久性的核心系统,其技术内涵远不止于悬架结构。从行驶系、转向系到制动系的协同工作,底盘调校是一门融合机械原理与工程实践的艺术。在电动化与智能化浪潮下,底盘技术正经历从被动适应到主动响应的变革,CDC可变阻尼减震器和空气悬架等技术的应用,让底盘具备了通过OTA升级实现性能优化的可能。国产汽车品牌如奇瑞、吉利等,通过平台化架构研发与调校经验积累,正推动底盘技术从“能用”向“好用”乃至“塑造独特驾驶风格”迈进。理解底盘技术,需超越简单的结构对比,关注其在实际路况中的滤震性、支

weixin_34413103的博客 387

AI计算,为什么要用GPU?

几年后,2012年,神经网络之父杰弗里·辛顿(Geoffrey Hinton)的两个学生——亚历克斯·克里切夫斯基(Alex Krizhevsky)、伊利亚·苏茨克沃(Ilya Sutskever),利用“深度学习+GPU”的方案,提出了深度神经网络AlexNet,将识别成功率从74%提升到85%,一举赢得Image Net挑战赛的冠军。根据形态,GPU可分为独立GPU(dGPU,discrete/dedicated GPU)和集成GPU(iGPU,integrated GPU),也就是常说的独显、集显。

经营范围: CTA、SRRC、CCC、移动、电信、联通入库 1678
上一篇: CP3405 CP2408 Lean UX
下一篇: MSDS 490: Healthcare Analytics
wx____codinghelp
博客等级 码龄2年 771粉丝 115原创
评论
成就一亿技术人!
拼手气红包6.0元
还能输入1000个字符
 
 条评论被折叠 查看
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付:10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值