ml 系统_需要保护您的ml系统的5个原因

ml 系统

We fuel our ambitions with our hard work and persistence every day to make our lives easier and convenient. That’s how humanity proceeds right! Spiderman is truly a visionary when he says “with great power, comes great responsibility”.

我们每天的辛勤工作和坚持不懈的努力使我们的雄心壮志使我们的生活更加轻松便捷。 这就是人类前进的方式！ 蜘蛛侠说“ 能力强大，责任重大 ” ，确实是一位有远见的人。

Machine Learning is one such power that boosts our convenience from Spotify’s suggestions based on our previous playlists to filtering spam and phishing emails. Though ML is an ingenious gift of advanced technology to us, it always remains in the ring succumbed by notorious malware and attacks.

机器学习就是一种强大的功能，它使我们从以前基于播放列表的Spotify的建议到过滤垃圾邮件和网络钓鱼电子邮件的便利性增加。 虽然ML是我们对先进技术的独到恩赐，但它始终被臭名昭著的恶意软件和攻击所笼罩。

Every business develops with the Trust of its customers and investors. This trust remains in place only if the customer’s data is not compromised and privacy is maintained. As we know, this has now become an uphill task with Data being the new fuel. Every organization big or small is racing against time to protect themselves and their clients from data breaches.

每一项业务都是在客户和投资者的信任下发展的。 仅当客户的数据没有受到损害并且保持隐私时，这种信任才会保留。 众所周知，随着Data成为新的燃料 ，这已经成为一项艰巨的任务。 每个大小的组织都在与时间赛跑，以保护自己和客户免遭数据泄露。

In this article, I will tell you the 5 good reasons why it is high time to protect your ML systems diligently.

在本文中，我将向您介绍5个充分的理由，说明该是时候该勤奋地保护您的ML系统了。

1. Data Theft: We expect eCommerce to surpass the trillion-dollar mark and obviously leading to a booming increase in digital frauds too. One such fraud is card-not-present fraud (CNP) which Juniper Research report says is expected to reach $130 billion by 2023. Another alarming news is that almost fifty percent of the eCommerce companies have fallen victim to it already.

1.数据盗窃：我们预计电子商务将突破万亿美元大关，并且显然也将导致数字欺诈的迅猛增长。 Juniper Research报告称，这种欺诈是卡不存在欺诈(CNP)，预计到2023年将达到1300亿美元。另一个令人震惊的消息是，近50％的电子商务公司已经成为其受害者。

2. Intellectual Property Theft: A patent is like a “Legal Monopoly” for your creativity and virtuoso (20 years by US Patent and Trademark Laws). In your corporate environment, the IP war may get worse than what is between Eminem and Iggy Azalea. These are the important forms of IP breaches seen in various sectors viz;

2.知识产权盗窃：一项专利就像是您的创造力和技巧的“法律垄断”(美国专利和商标法已有20年的历史)。 在您的公司环境中，IP战争可能比Eminem和Iggy Azalea之间的战争更糟。 这些是在各个部门看到的重要的IP违规形式。

a. Tech: Employees stealing computing solutions or outsourcing their work to make double paychecks.

一个。 技术：员工窃取计算解决方案或将工作外包以赚取双倍薪水。

b. Healthcare: Healthcare workers and pharmacists billing drugs beyond their authorization capacity.

b。 医疗保健：医疗保健工作者和药剂师在超出其授权能力的范围内对药物计费。

c. Education: Someone requesting for fake student loan grants.

C。 教育：有人要求提供虚假的学生贷款赠款。

d. Finance: Creation and Execution of fraudulent invoices of fake transactions.

d。 财务：创建和执行虚假交易的欺诈性发票。

3. Non-Compliance: Cyber Security laws in play are nothing less than corporate liabilities that come with a lot of checks and balances vis-à-vis.

3.不合规：网络安全法律正在发挥作用，无非是企业责任，而企业责任承担着许多制衡。

a. Full verbatim of Privacy Policy: According to the California Online Privacy Protection Act, 2004 every company which runs an online business and website or collects customer’s information (tracking or research purposes) residing in California (now covers almost the entire USA) must detail about its Privacy Policy eloquently in the website. This law explains the extent of the information collected, its purpose shared by a third party

一个。 隐私权政策的完整字词 ：根据2004年的《加利福尼亚州在线隐私保护法》，每家在加利福尼亚州(现在几乎覆盖整个美国)经营在线业务和网站或收集客户信息(跟踪或研究目的)的公司都必须详细说明其隐私政策网站上雄辩地声明了隐私政策。 该法律解释了所收集信息的范围，其目的由第三方共享

b. Personal data in all terms is very different than Personal Information or Personal Identifiable Information (you might have referred to them interchangeably, but each has a specific legal classification) and even the Federal Trade Commission (FTC) needs to propose its privacy laws and stand by it.

b。 个人数据在所有方面都与个人信息或个人身份信息有很大的不同(您可能会互换使用，但每个人都有特定的法律分类)，甚至联邦贸易委员会(FTC)都需要提出其隐私法并予以支持它。

c. Planning to spread your horizon outside the US. Great! But remember to book an appointment with your Lawyer. Data privacy laws outside the US are not parallel to US laws such as the EU’s GDPR bestows its citizens the right to collect, delete their information obtained by companies; while China’s data protection laws allow the Chinese government to interfere and save a copy of the source code of all software packages as well.

C。 计划将您的视野扩大到美国以外。 大！ 但请记住与律师预约。 美国境外的数据隐私法与美国法律并不平行，例如欧盟的GDPR授予其公民收集，删除公司获得的信息的权利； 而中国的数据保护法允许中国政府干预并保存所有软件包的源代码副本。

4. Degraded QoS: Quality of Service sheds light on the complete performance of the computing service using which an attacker can easily transmit huge amounts of optimized grants in the system. By doing so, the attacker can decrease the QoS or enforce upturn to the human operators. A (QoS)-enabled network becomes vulnerable to another form of attack- the QoS attack. The QoS-enabled network, such as a differentiated services network, provides distinct classes of service at varying costs. This difference in the charging rates may incite a few users to steal the bandwidth or any other network-related resource.

4.降级的QoS：服务质量阐明了计算服务的完整性能，攻击者可利用该服务轻松地在系统中传输大量优化的授权。 通过这样做，攻击者可以降低QoS或对操作人员实施上调。 启用(QoS)的网络容易受到另一种形式的攻击-QoS攻击。 支持QoS的网络(例如差异服务网络)以不同的成本提供不同的服务类别。 收费率的这种差异可能会刺激一些用户窃取带宽或任何其他与网络相关的资源。

5. Analytics Poisoning: Data poisoning is the most potential threat vector in the financial industry, as it can pull off any AI-based security system breach. It is highly manipulative and goes unnoticed for eons while continually harming the data and breaking the trust of your clients. Forensic experts also find it extremely difficult to trace such a sophisticated yet vulnerable state vector, indicating that cloud-based infrastructure requires skilled security modules.

5.分析中毒：数据中毒是金融行业中最潜在的威胁媒介，因为它可以实现任何基于AI的安全系统漏洞。 它具有高度的操纵性，并且在不断损害数据并破坏客户信任的同时，也引起了人们的注意。 法医专家还发现，追踪这样一个复杂而易受攻击的状态向量极为困难，这表明基于云的基础架构需要熟练的安全模块。

In addition to good ML services, you need security which always guards and filters out the potential threats, kind of like a modern data gladiator. I hope you are now more aware of the urgent need for securing your ML systems. Good Luck!

除了良好的ML服务，您还需要安全性，该安全性始终可以保护和过滤潜在的威胁，就像现代的数据角斗士一样。 我希望您现在更加意识到保护机器学习系统的迫切需求。 祝好运！

Learn more about what we are building at Scanta here

在这里了解更多关于我们在Scanta构建的内容

翻译自: https://towardsdatascience.com/5-reasons-why-your-ml-system-needs-to-be-protected-b2f5ea0c0636

ml 系统