2021.6.14-6.21 人工智能行业每周技术动态

上周五,从无锡飞到青岛,参加极视角举办的人工智能大会。

整个会议举办的还是比较成功的,上午主要是发布应用赛题各个主题演讲

下午是产业与应用论坛,和技术教育论坛的两个分会场。

涉及的面很广,虽然技术方面涉及的不多,但是多了解一些不同学者的看法,也很有收获。

从青岛回到无锡,周六周日两天,将他们公司的商业架构产品架构,也都整体的梳理了一下,感觉还是蛮有意思的。

传统方式的AI公司,一般是主抓几个领域的项目,将行业技术做深。

而他们走的是更广,平台的路线,通过合伙人/政府合作的方式,承接各类项目。

并将通用的一些项目难题,发布成开发者竞赛,或者将不同项目的需求,发布到平台,采用众包的方式,对接任务,进行训练。

这样有很大的好处:

(1)一方面可以对接各类项目,因为有大量的众包开发者。

(2)另一方面,走AI平台的概念,比做单一项目的团队,势能更大。

当然,还是很多政府相关的辅助,这里大白不多说。

不过,既然走广,也会存在弊端。

AI项目很多都是非标项目,算法只是其中一个模块。

如何梳理项目点?如何交付项目?都需要很多的人力物力。

因此,如何既走深,又走广,也是一个非常大的难题。

不过还是挺佩服他们的一些方式:

(1)执行力很高,比如从4月份会议立项,5月份开始准备,到举办,时间效率还是挺高的。

(2)营销方式也很厉害,比如从公共流量平台(知乎,B站),私域流量平台(公众号、极市开发者社区、微信群)、城市合伙人,各个渠道,互相影响。

不过从现场的会议感受,觉得还有一些可以改进的:

(1)如何建立参会者之间的联系,比如是否可以提前一天,举办AI技术夜话,分为各种主题,比如智能安防、互联网等,让用户可以互相交流。

(2)现场礼包中,缺少了一支笔,旁边的大姐想记点东西,找了身边好多人问有没有笔。

不过总体而言,大白觉得还是有不少收获的,也希望未来极视角可以越办越好。

之前的周报,大白会将每周的精华内容汇总起来,整理到《大白AI周报精华汇总》中,点击即可查看

后期需要哪方面的项目知识,可以直接去**对应阅读**。

大白也在不断收集更新各个项目算法作者及从业经验的视频分享,希望能让大家提高一些探索的效率,点击查看

整理汇总:江大白
内容周期:2021.6.14-6.21
同步公众号:江大白

目录

目录

 1  整理涉及公众号名单

2  行业精华文章汇总

2.1 基础知识方面

2.1.1 深度学习相关

2.2 研究方向方面

2.2.1 3D视觉

2.2.2 AI上色

2.2.3 自动驾驶

2.3 Opencv方面

3 行业&拓展阅读动态

3.1 行业动态

3.2 拓展阅读


 1  整理涉及公众号名单

(1)我爱计算机视觉
(2)Cver
(3)Datawhale
(4)量子位
(5)极市平台
(6)新智元
(7)机器之心
(8)AI算法与图像处理
(9)Opencv学堂
(10)PaperWeekly
(11)机器学习算法工程师
(12)AI研习社
(13)GiantPandaCV
(14)AI深度学习视线
(15)七月在线实验室
(16)人工智能前沿讲习
(17)AI科技评论
(18)机器学习算法与Python精研
(19)AIZOO
(20)微软研究员AI头条
(21)VALSE
(22)AI算法修炼营
(23)有三AI
(24)AlWalker
(25)AI公园
(26)AI人工智能初学者
(27)计算机视觉之路
(28)小白学视觉
(29)HyperAI超神经
(30)集智书童
(31)计算机视觉life

2  行业精华文章汇总

2.1 基础知识方面

2.1.1 深度学习相关

(1)终于有人总结了图神经网络!

链接:https://mp.weixin.qq.com/s/HDXfbP7jZp3qONKFcmrTbA

(2)ResNet也能用在3D模型上了!清华计图首创三角网格面片上的卷积神经网络:SubdivNet

链接:https://mp.weixin.qq.com/s/brqlqmnI1c1MYpY_8zaDKw

2.2 研究方向方面

2.2.1 3D视觉

(1)奥比中光CTO肖振中率队厦大开讲,详解3D视觉感知底层技术与产业应用

链接:https://mp.weixin.qq.com/s/jInF_0vdyCCn-eY9zt_fRg

(2)朱松纯:三维视觉是通用AI的关键,要以任务为导向

链接:https://mp.weixin.qq.com/s/HUl9Kr422xZVbb3Tf1MHcQ

(3)一部手机+几行代码,如何搞定三维重建?

链接:https://mp.weixin.qq.com/s/uYuwY7kwnC2Jx7eI7kWoFA

2.2.2 AI上色

(1)几行代码实现老照片上色复原!

链接:https://mp.weixin.qq.com/s/Zl304LQPTrisolYvOJlHIg

2.2.3 自动驾驶

(1)CVPR自动驾驶运动预测挑战赛:轻舟智航夺冠方案

链接:https://mp.weixin.qq.com/s/Olt7UkxSvexiZaoDl5A6YA

2.3 Opencv方面

(1)使用Python+OpenCV+Keras实现基于车牌的无口罩车辆驾驶员的惩罚生成

链接:https://mp.weixin.qq.com/s/NboBW3h3tbpcxJGfL_AIRg

(2)手动搭建一个车牌识别系统 | 附源码

链接:https://mp.weixin.qq.com/s/vndhjthaOCgqqNBSriwztQ

3 行业&拓展阅读动态

3.1 行业动态

(1)想Get热搜同款?GitHub开源神器让父亲重返18岁!

链接:https://mp.weixin.qq.com/s/Wv1pCENMq9tUgef_RlQF9A

(2)AI漫画纪元之崛起

链接:https://mp.weixin.qq.com/s/4h3J3JIHvrs7s9jBbjEFGA

(3)快手Y-tech:短视频智能创作的CV技术和发展趋势

链接:https://mp.weixin.qq.com/s/z_rid1Tf9xsTSTLyLakogQ

(4)他们翻遍用AI检测新冠的论文,一篇临床可用的也没有?!

链接:https://mp.weixin.qq.com/s/RhvgFa7hru845Bwh9jq5mQ

(5)恕我直言,很多小样本学习的工作就是不切实际的

链接:https://mp.weixin.qq.com/s/70r_6hIPmzyqxVQqunKqLQ

(6)跳舞手脚不协调?没关系,微视用AI打造你我的舞林大会,一张照片就可以

链接:https://mp.weixin.qq.com/s/Xw83qSLH89UkwOy1XKHEIQ

(7)这群云南象红遍地球 ,AI「扫地象」又凭啥横扫中国16城?

链接:https://mp.weixin.qq.com/s/HcaQwODWiFYRPKQCBZ4qew

3.2 拓展阅读

(1)引用次数在15000次以上的都是什么神仙论文?

链接:https://mp.weixin.qq.com/s/VpuGZAp7cvIFlOzzdJdbbw

(2)我就要朝9晚7!芯片大神带头反内卷,成了特斯拉最懒的人

链接:https://mp.weixin.qq.com/s/zX7pG2E4hrz31fa6X6Xtxg

(3)苹果前工程师5000字自述:乔布斯是怎么面试我的?

链接:https://mp.weixin.qq.com/s/pcuDP-6eCmNbP7d8bCqCLg

(4)人均估值5000万RMB,53岁程序员能做到的,你也能!

链接:https://mp.weixin.qq.com/s/p0EeWc3a1HIhLP5Uwrzq4Q

(5)谷歌最新薪资曝光:研究岗年薪最高达200多万,但远逊于人事主管

链接:https://mp.weixin.qq.com/s/p3k58CdTF2QZpscuA6r6-w

(6)我是一名AI视频 up 主,日更万部:这是我对人类世界的理解

链接:https://mp.weixin.qq.com/s/4KfOn7Aei2hhAf2HTU4oVA

(7)中国科学家成功让公鼠怀孕产子!全球首次,10只幼崽非常健康

链接:https://mp.weixin.qq.com/s/fke9PI_lr8YWuClLo6Pfvg

 

 

我想把我之前的结果都变成word形式生成,以及我做了太多结果,那些回归相关,我已经搞晕了。我把最后结果发给你,请你帮我在每个相关或回归前写一个表头,表明这是什么。 ___ ____ ____ ____ ____ © /__ / ____/ / ____/ 17.0 ___/ / /___/ / /___/ MP—Parallel Edition Statistics and Data Science Copyright 1985-2021 StataCorp LLC StataCorp 4905 Lakeway Drive College Station, Texas 77845 USA 800-STATA-PC https://www.stata.com 979-696-4600 stata@stata.com Stata license: Single-user 2-core perpetual Serial number: 501806366048 Licensed to: aaa bbb Notes: 1. Unicode is supported; see help unicode_advice. 2. More than 2 billion observations are allowed; see help obs_advice. 3. Maximum number of variables is set to 5,000; see help set_maxvar. . do "C:\Users\lenovo\AppData\Local\Temp\STD924_000000.tmp" . * =========================================================================== . * 验证 AI 投入是否具有滞后正向效应(终极修复版,确保一遍过) . * 作者:AI 助手 | 日期:2025年4月5日 . * =========================================================================== . . * --- 1. 清空内存,设置路径 --- . clear all . cd "D:\Users\lenovo\Desktop" // 修改为你自己的路径 D:\Users\lenovo\Desktop . . * --- 2. 加载主财务数据 --- . use "financial_cleaned.dta", clear . . * 检查是否有 year 变量 . capture confirm variable year . if _rc != 0 { . display as error "❌ 错误:数据中没有 'year' 变量!请检查数据结构" . exit 111 . } . . * --- 3. 检查 ai_annual.dta 是否存在并合并 --- . capture use "ai_annual.dta", clear . if _rc != 0 { . display as error "❌ 错误:无法打开文件 ai_annual.dta,请确认文件存在于桌面且名称正确" . exit 601 . } . keep year ai_mentions . save "ai_temp.dta", replace (file ai_temp.dta not found) file ai_temp.dta saved . . * 回到主数据 . use "financial_cleaned.dta", clear . merge m:1 year using "ai_temp.dta", nogen Result Number of obs ----------------------------------------- Not matched 0 Matched 39 ----------------------------------------- . . * --- 4. 检查 rd_annual.dta 是否存在并合并 --- . capture use "rd_annual.dta", clear . if _rc != 0 { . display as error "❌ 错误:无法打开文件 rd_annual.dta,请确认文件存在于桌面且名称正确" . exit 601 . } . keep year rd_expense . save "rd_temp.dta", replace (file rd_temp.dta not found) file rd_temp.dta saved . . * 回到主数据 . use "financial_cleaned.dta", clear . merge m:1 year using "ai_temp.dta", nogen Result Number of obs ----------------------------------------- Not matched 0 Matched 39 ----------------------------------------- . merge m:1 year using "rd_temp.dta", nogen Result Number of obs ----------------------------------------- Not matched 0 Matched 39 ----------------------------------------- . . * 删除临时文件 . erase "ai_temp.dta" . erase "rd_temp.dta" . . * --- 5. 按年聚合:将季度数据转为年度平均值 --- . collapse (mean) roa roe debt_ratio ln_revenue /// // 财务指标取均值 > (first) ai_mentions rd_expense, by(year) // AI 和 R&D 取当年值 . . * --- 6. 排序 --- . sort year . . * --- 7. 生成下一年 ROA --- . gen next_roa = roa[_n+1] if year[_n+1] == year + 1 (1 missing value generated) . label var next_roa "Next Year's ROA" . . * --- 8. 描述性统计 --- . summarize roa next_roa ai_mentions rd_expense ln_revenue debt_ratio Variable | Obs Mean Std. dev. Min Max -------------+--------------------------------------------------------- roa | 10 .0949047 .026076 .0620595 .1218738 next_roa | 9 .0919443 .0258139 .0620595 .1218738 ai_mentions | 10 52.9 56.58906 4 172 rd_expense | 10 2.74e+09 5.35e+08 1.73e+09 3.44e+09 ln_revenue | 10 23.12686 .3110618 22.51417 23.41911 -------------+--------------------------------------------------------- debt_ratio | 10 .0949394 .0146112 .0629635 .1099355 . . * --- 9. 相关性分析 --- . correlate next_roa ai_mentions rd_expense ln_revenue debt_ratio (obs=9) | next_roa ai_men~s rd_exp~e ln_rev~e debt_r~o -------------+--------------------------------------------- next_roa | 1.0000 ai_mentions | -0.7280 1.0000 rd_expense | 0.8877 -0.5749 1.0000 ln_revenue | -0.8232 0.6087 -0.5741 1.0000 debt_ratio | 0.4942 -0.7731 0.4171 -0.2729 1.0000 . . * --- 10. 回归分析 --- . regress next_roa c.ai_mentions c.rd_expense c.ln_revenue c.debt_ratio i.year, robust note: 2021.year omitted because of collinearity. note: 2022.year omitted because of collinearity. note: 2023.year omitted because of collinearity. note: 2024.year omitted because of collinearity. Linear regression Number of obs = 9 F(0, 0) = . Prob > F = . R-squared = 1.0000 Root MSE = 0 ------------------------------------------------------------------------------ | Robust next_roa | Coefficient std. err. t P>|t| [95% conf. interval] -------------+---------------------------------------------------------------- ai_mentions | 3.34e-06 . . . . . rd_expense | 1.73e-11 . . . . . ln_revenue | -.0532505 . . . . . debt_ratio | -.0980677 . . . . . | year | 2017 | .0035353 . . . . . 2018 | .0201439 . . . . . 2019 | .0230021 . . . . . 2020 | .0161762 . . . . . 2021 | 0 (omitted) 2022 | 0 (omitted) 2023 | 0 (omitted) 2024 | 0 (omitted) | _cons | 1.278377 . . . . . ------------------------------------------------------------------------------ . . * --- 11. 输出结果到 Word --- . * 如果未安装 outreg2,请先运行:ssc install outreg2, replace . capture which outreg2 . if _rc == 111 { . display as text "⚠️ 未安装 outreg2,正在尝试安装..." . ssc install outreg2, replace . } . outreg2 using ai_lag_effect.doc, replace ctitle("Future ROA") auto(2) /// > title("Table 1: Effect of AI Mentions on Next-Year ROA") ai_lag_effect.doc dir : seeout . . * --- 12. 绘制散点图:AI 提及 vs 下一年 ROA --- . twoway (scatter next_roa ai_mentions, mlabel(year) mcolor(red) msymbol(circle)) /// > (lfit next_roa ai_mentions, lcolor(blue)), /// > title("AI Mentions Predict Future ROA?") /// > subtitle("Current AI Mentions vs Next Year's ROA") /// > xtitle("AI Mentions in Current Year") /// > ytitle("ROA in Next Year") /// > legend(off) /// > name(ai_vs_future_roa, replace) . graph export "ai_vs_future_roa.png", width(900) height(600) replace (file ai_vs_future_roa.png not found) file ai_vs_future_roa.png saved as PNG format . . * --- 13. 绘制趋势图:ROA 与 AI 提及随时间变化 --- . preserve . twoway (line roa year, lcolor(blue) lwidth(medium)) /// > (line ai_mentions year, lcolor(red) yaxis(2) lwidth(medium)), /// > title("Trend Over Time") /// > xtitle("Year") /// > ytitle("ROA", axis(1)) /// > ytitle("AI Mentions", axis(2)) /// > xlabel(2016(1)2025) /// > legend(order(1 "ROA" 2 "AI Mentions")) /// > name(roa_ai_trend, replace) . graph export "roa_ai_trend.png", width(900) height(600) replace (file roa_ai_trend.png not found) file roa_ai_trend.png saved as PNG format . restore . . * --- 14. 显示数据预览 --- . list year roa next_roa ai_mentions in 1/12, sepby(year) noobs observation numbers out of range r(198); end of do-file r(198); . do "C:\Users\lenovo\AppData\Local\Temp\STD924_000000.tmp" . * =========================================================================== . * 验证 AI 投入是否具有滞后正向效应(终极修复版,确保一遍过) . * 作者:AI 助手 | 日期:2025年4月5日 . * =========================================================================== . . * --- 1. 清空内存,设置路径 --- . clear all . cd "D:\Users\lenovo\Desktop" // 修改为你自己的路径 D:\Users\lenovo\Desktop . . * --- 2. 加载主财务数据 --- . use "financial_cleaned.dta", clear . . * 检查是否有 year 变量 . capture confirm variable year . if _rc != 0 { . display as error "❌ 错误:数据中没有 'year' 变量!请检查数据结构" . exit 111 . } . . * --- 3. 检查 ai_annual.dta 是否存在并合并 --- . capture use "ai_annual.dta", clear . if _rc != 0 { . display as error "❌ 错误:无法打开文件 ai_annual.dta,请确认文件存在于桌面且名称正确" . exit 601 . } . keep year ai_mentions . save "ai_temp.dta", replace (file ai_temp.dta not found) file ai_temp.dta saved . . * 回到主数据 . use "financial_cleaned.dta", clear . merge m:1 year using "ai_temp.dta", nogen Result Number of obs ----------------------------------------- Not matched 0 Matched 39 ----------------------------------------- . . * --- 4. 检查 rd_annual.dta 是否存在并合并 --- . capture use "rd_annual.dta", clear . if _rc != 0 { . display as error "❌ 错误:无法打开文件 rd_annual.dta,请确认文件存在于桌面且名称正确" . exit 601 . } . keep year rd_expense . save "rd_temp.dta", replace (file rd_temp.dta not found) file rd_temp.dta saved . . * 回到主数据 . use "financial_cleaned.dta", clear . merge m:1 year using "ai_temp.dta", nogen Result Number of obs ----------------------------------------- Not matched 0 Matched 39 ----------------------------------------- . merge m:1 year using "rd_temp.dta", nogen Result Number of obs ----------------------------------------- Not matched 0 Matched 39 ----------------------------------------- . . * 删除临时文件 . erase "ai_temp.dta" . erase "rd_temp.dta" . . * --- 5. 按年聚合:将季度数据转为年度平均值 --- . collapse (mean) roa roe debt_ratio ln_revenue /// // 财务指标取均值 > (first) ai_mentions rd_expense, by(year) // AI 和 R&D 取当年值 . . * --- 6. 排序 --- . sort year . . * --- 7. 生成下一年 ROA --- . gen next_roa = roa[_n+1] if year[_n+1] == year + 1 (1 missing value generated) . label var next_roa "Next Year's ROA" . . * --- 8. 描述性统计 --- . summarize roa next_roa ai_mentions rd_expense ln_revenue debt_ratio Variable | Obs Mean Std. dev. Min Max -------------+--------------------------------------------------------- roa | 10 .0949047 .026076 .0620595 .1218738 next_roa | 9 .0919443 .0258139 .0620595 .1218738 ai_mentions | 10 52.9 56.58906 4 172 rd_expense | 10 2.74e+09 5.35e+08 1.73e+09 3.44e+09 ln_revenue | 10 23.12686 .3110618 22.51417 23.41911 -------------+--------------------------------------------------------- debt_ratio | 10 .0949394 .0146112 .0629635 .1099355 . . * --- 9. 相关性分析 --- . correlate next_roa ai_mentions rd_expense ln_revenue debt_ratio (obs=9) | next_roa ai_men~s rd_exp~e ln_rev~e debt_r~o -------------+--------------------------------------------- next_roa | 1.0000 ai_mentions | -0.7280 1.0000 rd_expense | 0.8877 -0.5749 1.0000 ln_revenue | -0.8232 0.6087 -0.5741 1.0000 debt_ratio | 0.4942 -0.7731 0.4171 -0.2729 1.0000 . . * --- 10. 回归分析(修正版:避免过拟合)--- . * 不再使用 i.year,改用时间趋势或仅控制协变量 . . * 方法一:简单稳健回归(推荐小样本) . regress next_roa c.ai_mentions c.rd_expense c.ln_revenue c.debt_ratio, robust Linear regression Number of obs = 9 F(3, 4) = . Prob > F = . R-squared = 0.9505 Root MSE = .00812 ------------------------------------------------------------------------------ | Robust next_roa | Coefficient std. err. t P>|t| [95% conf. interval] -------------+---------------------------------------------------------------- ai_mentions | -.0000538 .000124 -0.43 0.687 -.000398 .0002905 rd_expense | 2.87e-11 7.17e-12 4.00 0.016 8.79e-12 4.86e-11 ln_revenue | -.0346581 .0132604 -2.61 0.059 -.0714749 .0021587 debt_ratio | .1340701 .1677207 0.80 0.469 -.3315972 .5997375 _cons | .8053355 .2946586 2.73 0.052 -.0127679 1.623439 ------------------------------------------------------------------------------ . . * 方法二:加入时间趋势(可选) . * gen time_trend = year - 2016 . * regress next_roa c.ai_mentions c.rd_expense c.ln_revenue c.debt_ratio c.time_trend, robust . . * --- 11. 输出结果到 Word --- . outreg2 using ai_lag_effect.doc, replace ctitle("Future ROA") auto(2) /// > title("Table 1: Effect of AI Mentions on Next-Year ROA") ai_lag_effect.doc dir : seeout . . * --- 12. 绘制散点图 --- . twoway (scatter next_roa ai_mentions, mlabel(year) mcolor(red) msymbol(circle)) /// > (lfit next_roa ai_mentions, lcolor(blue)), /// > title("AI Mentions Predict Future ROA?") /// > subtitle("Current AI Mentions vs Next Year's ROA") /// > xtitle("AI Mentions in Current Year") /// > ytitle("ROA in Next Year") /// > legend(off) /// > name(ai_vs_future_roa, replace) . graph export "ai_vs_future_roa.png", width(900) height(600) replace file ai_vs_future_roa.png saved as PNG format . . * --- 13. 绘制趋势图 --- . preserve . twoway (line roa year, lcolor(blue) lwidth(medium)) /// > (line ai_mentions year, lcolor(red) yaxis(2) lwidth(medium)), /// > title("Trend Over Time") /// > xtitle("Year") /// > ytitle("ROA", axis(1)) /// > ytitle("AI Mentions", axis(2)) /// > xlabel(2016(1)2025) /// > legend(order(1 "ROA" 2 "AI Mentions")) /// > name(roa_ai_trend, replace) . graph export "roa_ai_trend.png", width(900) height(600) replace file roa_ai_trend.png saved as PNG format . restore . . * --- 14. 显示数据预览(修正范围)--- . list year roa next_roa ai_mentions in 1/10, sepby(year) noobs +--------------------------------------+ | year roa next_roa ai_men~s | |--------------------------------------| | 2016 .121548 .1202005 4 | |--------------------------------------| | 2017 .1202 .1179928 5 | |--------------------------------------| | 2018 .117993 .1218738 13 | |--------------------------------------| | 2019 .121874 .1115533 16 | |--------------------------------------| | 2020 .111553 .0837875 20 | |--------------------------------------| | 2021 .083788 .0620595 32 | |--------------------------------------| | 2022 .062059 .0654863 61 | |--------------------------------------| | 2023 .065486 .0806397 89 | |--------------------------------------| | 2024 .08064 .063905 117 | |--------------------------------------| | 2025 .063905 . 172 | +--------------------------------------+ . . * --- 15. 成功提示 --- . display as text "✅ 分析完成!" ✅ 分析完成! . display as text "📊 回归结果已保存至:ai_lag_effect.doc" 📊 回归结果已保存至:ai_lag_effect.doc . display as text "🖼️ 图表已保存至桌面:" 🖼️ 图表已保存至桌面: . display as text " • ai_vs_future_roa.png" • ai_vs_future_roa.png . display as text " • roa_ai_trend.png" • roa_ai_trend.png . end of do-file .
最新发布
11-12
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
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
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