IEEE Technology and Society Magazine - March 2021 - 64

AI for a Fair, Just, and Equitable World

Table 1. Data sets (D) used for DR classification. It shows the
distribution from each public data set with different DR stages.

and Inception-v4 [70] (m1, m2, m3 classification)
architectures. Figure 3 represents the workflow of
the models.

Results
Table 2 presents the accuracy, loss, specificity,
and sensitivity of the test sets evaluation over d1, d 2,
and d3. The accuracy of d 1/m 0 is very high; however,
this is a segmentation-prepossessing task. The accuracy for d1/m 1 has the best classification results. Taking into account that a trained professional grader
reaches a 91% for specificity and 99% for sensitivity
[71], we consider this as a good result for transfer
learning. We believe that the performance of the
models could improve if the number of images

Figure 3. Proposed DR models m0 (U-net), m1, m2,
and m3 (Inception V4).

Table 2. Test accuracy, test loss, specificity, and
sensitivity results for the different data sets, where
D is the data and M is the model.

increases. The performance of the models with local
data set images (d4, d 5, d 6) decreases significantly.
Preliminary analysis of the image quality shows that
the local data set has lower quality. However, further
analysis and image quality evaluation has to be done.
This is commonly overlooked in DL systems, and
these systems are trained and tested on high-quality
image data sets [72]. These shortcomings could be
more dramatic for medical applications [72]. As for
the technical quality, it seems that the quality of the
results is moderate for some data sets.

Organization and strategies
So far, some important points learned by developing and implementing this pilot are the following:
* Development of the strategic plan: Our strategic plan takes into account the mission, vision,
political agenda, alliances, production structure, values-ethics, communication-dissemination tactics, and necessary resources to real
deployment.
* Data governance: The organizational structure of
the Coordination General de Innovation Gubernamental de la jefatura de gabinete del Estado
de Jalisco (CGI- JAL) facilitates data governance.
It is possible to get access to the data transversely
with many government departments. We preserve secure data integrity and anonymization,
ensuring that data ethics, privacy, and fundamental rights are the principles that guide the use
of a trustworthy AI in the LAC region.
* Prospect and select AI use cases: We take into
account the following points:
a) We take into account the public policies versus impact factor. We use an impact factor
matrix to prioritize the government agenda
and the possible solutions based on implementation effort and resources.

IEEE Technology and Society Magazine

https://innovacion.jalisco.gob.mx/ https://innovacion.jalisco.gob.mx/ https://plan.jalisco.gob.mx/

IEEE Technology and Society Magazine - March 2021

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - March 2021