Agile Network India - Gurugram
Title: Agile Adaptation: Driving Progress in Generative AI Projects by Sujata Bhutani
Date: 20th April 2024
Hosted by: The NorthCap University
5. SUJATA
Sujata is doctorate by education, Teacher by passion, and parent by
divine bliss.
She has 15+ years experience in industry and academia. A part of
teaching currently working as Research Advisory and Reviewer of
International organizations, journals and conferences respectively.
Sujata is certified tester and conscious parent with great appreciation
for the role the culture, environment and nurturers in life.
“Agile is not a model, it’s a mindset”.
Agile is a set of principles and values that don’t tell how to work but they focus on helping the team think
and interact in ways to achieve agility.
Agile values and principles not tell u what to do but help the team to decide what should do
.
Agile is a light-weight approach for enhancing modeling and documentation efforts for other software
processes such as XP and RUP
Agile
15. Generative Adversarial Network (GAN)
Generative : Generator/Creator/Producer
Its about producing/generating /creating From the name its that its
generative , it is able to produce any sensible data
Adversarial : Inspector/Checker/Opposition Party
Objective is to make fool to this component
N: Neural Network
16. Generative Adversarial Network (GAN)
It consists of two models that compete each other to analyze, capture and copy the variations
within a dataset
17. Cross Entropy Loss Function in GAN
It’s a min-max optimization formulation where the Generator wants to minimize the objective function whereas the
Discriminator wants to maximize the same objective function.
The Discriminator wants to drive the likelihood of D(G(z)) to 0. Hence it wants to maximize (1-D(G(z))) whereas the
Generator wants to force the likelihood of D(G(z)) to 1 so that Discriminator makes a mistake in calling out a
generated sample as real. Hence Generator wants to minimize (1-D(G(z)).
19. Cycle GAN
As mentioned earlier there are 2 kinds of functions being learned, one of them is G which transforms X to Y and the
other one is F which transforms Y to X and it comprises two individual GAN models. So, you will find 2 Discriminator
function Dx, Dy.
As part of Adversarial formulation, there is one Discriminator Dx that classifies whether the transformed Y is
indistinguishable from Y. Similarly, there is one more Discriminator Dy that classifies whether is indistinguishable
from X.
20. Text2image GAN
In this formulation, instead of giving only noise as input to the Generator, the textual description is first transformed into
a text embedding, concatenated with noise vector and then given as input to the Generator.
As an example, the textual description has been transformed into a 256-dimensional embedding and concatenated with
a 100-dimensional noise vector [which was sampled from a latent space which is usually a random Normal distribution].
22. Augmenting the workforce with AI
Scrum Masters/Team Coaches (SM/TC) can use AI to analyze flow data to identify and predict potential
bottlenecks and delays and suggest improvements, leading to more effective planning, delivery, and
retrospectives.
Product Owners can use AI to suggest enhancements to user stories and acceptance criteria, ensuring
the team constantly works on the most valuable backlog items.
Release Train Engineers (RTE) can leverage AI’s real-time data analysis and predictive modeling to
identify and address workflow inefficiencies, align capacity with demand, and streamline value delivery.
Lean Portfolio Management can leverage historical performance data and market trends to identify
strategic themes and opportunities, enabling more informed decisions on epic prioritization.
Additionally, AI-driven analytics can aid in establishing dynamic, data-driven lean budgets and ‘smart’
KPIs, ensuring resources are allocated efficiently and aligned with business objectives, thereby
enforcing lean budget guardrails more effectively.
Software engineers realize significant benefits from generative AI ‘copilot’ development tools that
assist in coding, automated testing, and code review processes. These tools can suggest optimal
coding practices, detect bugs early, and even propose fixes, streamlining the development cycle.
23.
24. Challenges in Generative AI Development
1. Complexity of Models: Generative AI models are often complex and
computationally intensive, requiring significant expertise in machine learning and
computer science.
2. Data Quality and Diversity: Generative AI models rely on large amounts of high-
quality data to produce meaningful outputs. Ensuring data quality and diversity is
essential for model performance.
3. Ethical Considerations: Generating realistic content raises ethical concerns, such
as the potential for misuse or bias. Ethical considerations must be carefully
addressed throughout the development process
25. The Role of Agile Adaptation
Agile AI Development: Implementing AI projects using agile principles and values can enhance flexibility,
adaptability, and efficiency in delivering AI solutions.
Continuous Feedback: Agile principles emphasize continuous feedback. This can be leveraged in AI
projects to refine models, improve performance, and meet evolving requirements.
Scrum for AI: Applying Scrum framework within AI development can help teams manage tasks, prioritize
work, and ensure regular progress updates in AI projects.
Sprints for AI Development: Breaking down AI projects into sprints allows for incremental development,
testing, and integration of AI models, aligning with agile principles.
Cross-functional Teams: Forming cross-functional teams with expertise in both AI and agile practices can
promote collaboration and communication, enhancing project outcomes.
26. Learning
Agile provides a robust framework for navigating the complexities of Generative AI
development, enabling teams to iterate rapidly, incorporate feedback, and adapt to
changing requirements and technologies.
As the field of Generative AI continues to evolve, Agile adaptation will play a vital
role in driving innovation and shaping the future of AI-powered creativity.