* Mdnotes File Name: [[khomhSoftwareEngineeringMachineLearning2018]]

# Green Annotations (18/12/2020, 00:14:17)

> "we still experience failures and shortcomings in the resulting soft­ ware systems. The main reason is the shift in the development paradigm in­ duced by ML and AI." ([Khomh et al 2018:81](zotero://open-pdf/library/items/5486JT7B?page=1))

> "with ML techniques, these rules are inferred from training data (from which the requirements are gener­ ated inductively)." ([Khomh et al 2018:81](zotero://open-pdf/library/items/5486JT7B?page=1))

> "This paradigm shift makes reasoning about the be­ havior of software systems with ML components difficult, resulting in software systems that are intrinsi­ cally challenging to test and verify." ([Khomh et al 2018:81](zotero://open-pdf/library/items/5486JT7B?page=1))

> "the learned behavior of an ML­based system might be incorrect, even if the learning algorithm is imple­ mented correctly, a situation in which traditional testing techniques are ineffective." ([Khomh et al 2018:81](zotero://open-pdf/library/items/5486JT7B?page=1))

> "critical problem is how to effectively develop, test, and evolve such systems, given that they don't have (complete) specifications or even source code corresponding to some of their critical behaviors." ([Khomh et al 2018:81](zotero://open-pdf/library/items/5486JT7B?page=1))

*TESTING (ML) SYSTEMS THAT LACK SPECIFICATIONS OR EVEN SOURCE CODE ([note on p.81](zotero://open-pdf/library/items/5486JT7B?page=1))*

 

> "AI technology's strength comes from the ability to abstract up from different factors of varia­ tion between environments, to obtain models that can general­ ize and transfer to situations that weren't encountered before" ([Khomh et al 2018:82](zotero://open-pdf/library/items/5486JT7B?page=2))

> "AI tech­ nologies' main challenge is" ([Khomh et al 2018:82](zotero://open-pdf/library/items/5486JT7B?page=2))

> "he need for sufficient, labeled data to cover all important factors (features) of a given problem." ([Khomh et al 2018:82](zotero://open-pdf/library/items/5486JT7B?page=2))

> "AI, in fact, needs more training data than humans do!" ([Khomh et al 2018:82](zotero://open-pdf/library/items/5486JT7B?page=2))

> "appli­ cations of AI still risk being limited to domains in which labeled data is cheap." ([Khomh et al 2018:82](zotero://open-pdf/library/items/5486JT7B?page=2))

> "instead of touting a "100 percent self­driving car," auto­ motive companies should advertise their products as "AI­assisted cars," with a clear list of the ways in which AI is assisting." ([Khomh et al 2018:82](zotero://open-pdf/library/items/5486JT7B?page=2))

> "If a traditional computer science algorithm can solve a problem, we should just use that." ([Khomh et al 2018:82](zotero://open-pdf/library/items/5486JT7B?page=2))

> "how can we perform adequate quality assurance (QA) of AI models, given that the number of environments in which the mod­ els will be deployed is unlimited and that the human operator will re­ quire a detailed explanation of any failures?" ([Khomh et al 2018:82](zotero://open-pdf/library/items/5486JT7B?page=2))

> "use AI tech­ nology to reduce the search space of the environments to be tested" ([Khomh et al 2018:82](zotero://open-pdf/library/items/5486JT7B?page=2))

> "AI impacts the hu­ mans' recommendations, those rec­ ommendations are also a human filter for AI failures." ([Khomh et al 2018:83](zotero://open-pdf/library/items/5486JT7B?page=3))

> "Creating an efficient syntax for automatic differentiation that can deliver ease of implementation, per­ formance, usability, and flexibility is important but difficult." ([Khomh et al 2018:83](zotero://open-pdf/library/items/5486JT7B?page=3))

*CHALLENGES ([note on p.83](zotero://open-pdf/library/items/5486JT7B?page=3))*

 

> "esting and debugging these implementations are also salient challenges." ([Khomh et al 2018:83](zotero://open-pdf/library/items/5486JT7B?page=3))

> "How should software develop­ ment teams integrate the AI model lifecycle (training, testing, deploying, evolving, and so on) into their software process?" ([Khomh et al 2018:84](zotero://open-pdf/library/items/5486JT7B?page=4))

> "What new roles, artifacts, and activities come into play, and how do they tie into existing agile or DevOps processes?" ([Khomh et al 2018:84](zotero://open-pdf/library/items/5486JT7B?page=4))

> "testing" ([Khomh et al 2018:84](zotero://open-pdf/library/items/5486JT7B?page=4))

> "intersections" ([Khomh et al 2018:84](zotero://open-pdf/library/items/5486JT7B?page=4))

> "critical challenges of as­ suring the quality of AI and software systems in general." ([Khomh et al 2018:84](zotero://open-pdf/library/items/5486JT7B?page=4))