Caraga AI Roadmapping Activity

April 7-8, 2025, at the Grand Palace Hotel, Butuan City

Engr. Eduardo D. Piedad, Jr

Content
1. AI in the Philippines & the ASEAN
• Ethical AI Development and Governance
• OneDOST4U! Providing Solutions, Opening Opportunities
for the AI Sector through STI
2. DOST-ASTI and its AI Initiatives
• AI projects
• Demos
Content
1. AI in the Philippines & the ASEAN
• Ethical AI Development and Governance
• OneDOST4U! Providing Solutions, Opening
Opportunities for the AI Sector through STI
2. DOST-ASTI and its AI Initiatives
• AI projects
• Demos
Microsoft and LinkedIn
2024 Work Trend Index Annual Report

From https://www.microsoft.com/en-us/worklab/work-trend-index/ai-at-work-is-here-now-comes-the-hard-part
Released May 8, 2024
Filipino Workers Ahead in
AI Adoption at Work

• 86% of Filipino knowledge

workers use AI,
surpassing, higher than
the global average of 75%
and regional average of
83%.
• 83% of Filipino AI users
bring their tools to work—
Bring Your Own AI
(BYOAI). Globally, 78% of
employees engage in
BYOAI, and regionally, 79%

From https://news.microsoft.com/en-ph/2024/05/23/microsoft-and-linkedin-release-2024-work-
trend-index-on-ai-use-at-work-in-the-philippines/?msockid=0dc505fce6eb69f91599167ce7766855
Organizations
Struggle with Training
and Strategy

• 55% of Filipino leaders

admit their
organization lacks a
clear plan or vision for
AI implementation,
and only 30% of
Filipino AI users
receive adequate
training.

From https://news.microsoft.com/en-ph/2024/05/23/microsoft-and-linkedin-release-2024-work-
trend-index-on-ai-use-at-work-in-the-philippines/?msockid=0dc505fce6eb69f91599167ce7766855
 The Emergence of Generative AI

Generative AI is
rapidly changing
the technological
and economic
landscape.

Prompt: A stylish woman walks down a Tokyo street filled with warm glowing neon and animated city
signage. She wears a black leather jacket, a long red dress, and black boots, and carries a black purse. She
wears sunglasses and red lipstick. She walks confidently and casually. The street is damp and reflective,
creating a mirror effect of the colorful lights. Many pedestrians walk about. From
https://openai.com/index/sora/
 How do we ensure that the public
and private sectors in the
Philippines can fully embrace AI
to drive productivity, workforce
engagement, and equitable
development, while addressing
the strategic, training, and
infrastructure gaps that hinder
effective adoption?
Photo from https://news.microsoft.com/en-ph/2024/05/23/microsoft-
and-linkedin-release-2024-work-trend-index-on-ai-use-at-work-in-
the-philippines/?msockid=0dc505fce6eb69f91599167ce7766855
Ethical and Social Considerations

Addressing Bias Ensuring Transparency Promoting Inclusive

and Fairness in AI: and Accountability: AI Development
It is crucial to identify and Transparency in AI systems means Inclusive AI development involves
mitigate biases in AI systems to providing clear, understandable creating AI technologies that serve
ensure they do not perpetuate or explanations of how AI decisions are diverse populations and addressing
amplify existing societal made. This fosters trust and allows for the digital divide. This requires
inequalities. This involves accountability, ensuring that AI concerted efforts to involve
implementing robust data developers and users can be held underrepresented groups in AI
governance practices and responsible for the impacts of AI development processes and
regularly auditing AI models for technologies ensuring equitable access to AI
discriminatory patterns benefits
 Principles and Guidelines for an Ethical and
Trustworthy Use of AI in the Philippine Government

The Department of Information and Communications Technology (DICT) and the Civil Service Commission (CSC) have drafted a Joint Memorandum Circular to
establish ethical guidelines for AI use in the Philippine government, focusing on principles like transparency, accountability, and human-centricity to ensure AI
technologies are deployed responsibly and enhance public service delivery. Retrieved from https://dict.gov.ph/ictstatistics/stakeholder-consultation-on-the-draft-joint-
memorandum-circular-on-the-principles-and-guidelines-for-an-ethical-and-trustworthy-use-of-artificial-intelligence-ai-in-the-government/
 Proposed Legislations, 19th Congress
Several AI-related bills are currently pending in the 19th
Session of the House of Representatives.
• House Bill 7913: Establishing a regulatory framework
for AI, including the creation of the Philippine Council on
Artificial Intelligence and defining prohibited acts.
https://issuances-library.senate.gov.ph/bills/house-bill-no-7913-19th-congress
• House Bill 7983: Creating a national strategy for AI
development and establishing the National Center for
Artificial Intelligence Research.
https://issuances-library.senate.gov.ph/bills/house-bill-no-7983-19th-congress
• House Bill 7396: Promoting the development and
regulation of AI in the Philippines.
https://issuances-library.senate.gov.ph/bills/house-bill-no-7396-19th-congress
• House Bill 9448: Regulating the use of AI and
automation in the labor industry. Stock photo
https://issuances-library.senate.gov.ph/bills/house-bill-no-9448-19th-congress
 Government AI Readiness Index 2023

https://oxfordinsights.com/ai-readiness/ai-readiness-index/
 Government AI Readiness Index 2023
Technology Sector Data and
Economy Total Score Government Pillar Global Rankings
Pillar Infrastructure

Singapore 81.97 90.40 66.19 89.32 2

Malaysia 68.71 79.99 54.13 72.00 23
Thailand 63.03 77.21 41.33 70.55 37
Indonesia 61.03 76.24 43.48 63.38 42
Philippines 51.98 65.43 34.38 56.13 65
Viet Nam 54.48 69.04 37.82 56.58 59
Brunei Darussalam 48.10 39.57 41.86 62.87 74
Lao PDR 33.05 28.39 25.81 44.96 136
Cambodia 31.88 27.93 22.53 45.17 145
Myanmar 30.91 21.63 27.98 43.14 149
Timor-Leste 29.77 24.24 21.32 43.75 157
Average Score 50.45 54.55 37.89 58.90

https://oxfordinsights.com/ai-readiness/ai-readiness-index/
 Critical AI Regulatory Blocks in ASEAN
Country Data Protection and Cybersecurity Copyright and Consumer Protection
Privacy Intellectual Property
Republic Act 8293
Cybercrime Prevention Act Electronic Commerce Act
Philippines Data Privacy Act (2012) (Intellectual Property
(2012) (2000)
Code) (1997)
Personal Data Protection Consumer Protection (Fair
Singapore Cybersecurity Act (2018) Copyright Act (2021)
Act (2012) Trading) Act (2003)
Consumer Protection Act
(1979), Civil and
Personal Data Protection
Thailand Cybersecurity Act (2019) Copyright Act (1994) Commercial Code Unfair
Act (2019)
Contract Terms Act (1997),
Penal Code
Law on Information Law on Cyber Information
Technology (2006), Law on Security (2015), Law on Intellectual Property Law Law on Protection of
Vietnam
Protection of Consumers' Network Information (2005) Consumers' Rights (2010)
Rights (2010) Security (2015)
Constitution of the Constitution of the
Law No. 8/2016 on
Timor-Leste Democratic Republic of No specific law exists. Democratic Republic of
Consumer Protection
Timor-Leste Timor-Leste

Updated from ISEAS, 2023, ASEAN’s New Dilemma: Managing the Artificial Intelligence (AI) Space.
ASEAN Guide on AI Governance and Ethics
 DEPARTMENT OF SCIENCE AND TECHNOLOGY

Ethical AI Development and

Governance
30 January 2025

www.dost.gov.ph osec@dost.gov.ph DOST Philippines | Providing Solutions, Opening Opportunities

 The Trolley
Problem

Image credit: Alexander_P/shutterstock.com

Retrieved from https://www.iflscience.com/a-youtuber-recreated-the-trolley-problem-in-real-life-73524

www.dost.gov.ph osec@dost.gov.ph DOST Philippines | Providing Solutions, Opening Opportunities

 Why Ethical AI
Matters

Moral Self-Consideration by an AI (Evil AI Cartoons). Image from

https://humancentered-ai.eu/lecture-intro-to-hcai/

www.dost.gov.ph osec@dost.gov.ph DOST Philippines | Providing Solutions, Opening Opportunities

 Collingridge
Dilemma

“When change is easy, the need for it

cannot be foreseen; when the need for
change is apparent, change has
become expensive, difficult, and time-
consuming.”
The Social Control of Technology by
David Collingridge (1980).

Image from https://demoshelsinki.fi/what-is-the-collingridge-

dilemma-tech-policy/

www.dost.gov.ph osec@dost.gov.ph DOST Philippines | Providing Solutions, Opening Opportunities

Key Ethical Challenges in AI

CNN to predict pneumonia from chest X-rays identifies specific and

general features, including non-diagnostic artifacts like technician
markers, which may inadvertently influence disease prediction

Bias and Fairness Transparency and Explainability Accountability

AI can perpetuate inequalities by Understanding AI decisions is Clear responsibility for AI-driven

unintentionally learning human essential for trust, especially in decisions fosters innovation and
prejudices from historical data. critical sectors like healthcare. public confidence.

www.dost.gov.ph osec@dost.gov.ph DOST Philippines | Providing Solutions, Opening Opportunities

Key Ethical Challenges in AI

Facial manipulation technologies: Identity Swap, using images from the

Facebook CEO Mark Zuckerberg testifying before the U.S. Congress in Celeb-DF database; and Expression Swap, featuring images from the 2024 Stanford AI Index: The rapid escalation in training costs, with models like GPT-4 and
2018, following the Cambridge Analytica scandal. Photo: Xinhua /
FaceForensics++ dataset (Tolosana et al., 2020). Gemini Ultra reaching $78 million and $191 million, respectively
Barcroft from The Guardian

Privacy and Data Protection Misinformation and Deepfakes Sustainability

AI must respect privacy, Generative AI risks amplifying Balancing AI’s benefits with its
safeguarding personal data misinformation, undermining environmental impact is a
against misuse and surveillance. truth, and eroding public trust. pressing ethical challenge.

www.dost.gov.ph osec@dost.gov.ph DOST Philippines | Providing Solutions, Opening Opportunities

 DOST’s Role & Initiatives

United Nation’s 68th session of the Commission on the Status of Women (CSW) Interactive ASEAN COSTI-85 Side Event: Enabling Environment for a Human-Centered Adoption of
Dialogue AI in ASEAN, 3 June 2024 in Siem Reap, Cambodia

www.dost.gov.ph osec@dost.gov.ph DOST Philippines | Providing Solutions, Opening Opportunities

Management
NIST AI Risk

Framework

Lifecycle and key dimensions of AI

systems, focusing on the
integration of Test, Evaluation,
Verification, and Validation
(TEVV) processes across the
lifecycle. This approach ensures
that AI systems are trustworthy,
contextually appropriate, and
aligned with ethical, legal, and
societal considerations.
Retrieved from
https://nvlpubs.nist.gov/nistpubs/
ai/nist.ai.100-1.pdf

www.dost.gov.ph osec@dost.gov.ph DOST Philippines | Providing Solutions, Opening Opportunities

 ASEAN Guide on AI Governance and Ethics
Guiding Principles

Transparency and Explainability Privacy and Data Governance

Ensuring stakeholders understand how AI systems make Respecting data privacy, protecting sensitive data, and
decisions and are aware of their use. adhering to data governance laws.

Fairness and Equity Accountability and Integrity

Mitigating bias and ensuring AI decisions do not Ensuring organizations remain responsible for AI decisions
discriminate against specific demographics. and comply with laws and ethical norms.

Security and Safety Robustness and Reliability

Protecting AI systems from malicious attacks and Designing AI systems to operate consistently and handle
ensuring they operate without causing harm. errors or unforeseen circumstances.

Human-centricity

Prioritizing human well-being and social good while

avoiding harm or exploitation.

www.dost.gov.ph osec@dost.gov.ph DOST Philippines | Providing Solutions, Opening Opportunities

 ASEAN AI
Governance
Internal Governance
Framework Structures and Measures Determining the Level of
Human Involvement

The ASEAN AI Governance Framework

focuses on four key areas to guide
organizations through the responsible
Privacy and Data Accountability and
use of AI.
Governance Integrity
Retrieved from
https://asean.org/book/asean-guide-on-
ai-governance-and-ethics/

www.dost.gov.ph osec@dost.gov.ph DOST Philippines | Providing Solutions, Opening Opportunities

 Data Protection Cybersecurity
and Privacy

Critical Regulatory Cybercrime Prevention Act

Data Privacy Act (2012) (2012)
Blocks for
Copyright and Consumer

Responsible and Intellectual

Property
Protection

Ethical AI Republic Act 8293

(Intellectual Property Code) Electronic Commerce Act
(1997) (2000)

www.dost.gov.ph osec@dost.gov.ph DOST Philippines | Providing Solutions, Opening Opportunities

www.dost.gov.ph osec@dost.gov.ph DOST Philippines | Providing Solutions, Opening Opportunities
OneDOST4U!
Providing Solutions,
Opening Opportunities
for the AI Sector
through STI
AI has the potential to add $1 trillion SEA GDP by 2030

Source: EDBI and Kearney (2020)

The Future of AI Industry in the Philippines
 DOST AI R&D Support 2018 – 2023 MV1

Artificial
Intelligence
1%

28 298M
Projects Budget
Slide 31

MV1 Hi, Ate [@Audrey Jane]. For updating, please.

Mike Vertudazo, 2025-01-21T10:01:03.711
MV1

DOST AI R&D Framework

A 10-year Framework for Artificial Intelligence
R&D in PHL (DOST-PCIEERD)
AI Landscape
for 2019-2029 •
•
Data-driven AI applications are beneficial to stakeholders
Academe as the main partner of the government in R&D
• Industry focuses on practical solutions using their own resources
Key Components Complementing the National AI • Fragmented funding of AI programs and projects

Strategy: Aspirations
1. Universal and affordable access to AI infrastructure

1. National Infrastructure 2. Upskilling the workforce

3. Science-based solutions for socio-economic
• AI R&D Centers with HPCs opportunities
4. Unprecedented innovation for the industry sector
5. Enhanced Policy Support and Stakeholder
• Mission-driven AI R&D Centers Engagement

2. Talent and Skills Development AI R&D Priorities

• Broadening curriculum on AI
• Recruitment and training programs
• Training through R&D
3. Research and Data
4. Policies and Stakeholder Engagement
Slide 32

MV1 https://pcieerd-my.sharepoint.com/:p:/g/personal/cgreyes_pcieerd_dost_gov_ph/EZq0V6kPJQZHi3mxQ4na7bsBviNVR-lkB5WJ4oHkDXKAcg?e=1K6DHA
Mike Vertudazo, 2023-04-27T08:13:52.062
 Upskilling the Workforce

49,391
3 AI Trainings with SPARTA Scholars
(From 2020-present)

143 participants from 37 HEIs

COMPLETED

AI on Cross-cutting Applications/Platforms
Project 1: Automated Labeling Machine – Project 2: ASTI Automated Labeling
Large Scale Initiative (ALaM-LSI) Machine (ASTI-ALaM)
Project Leader: Dr. Jose Ildefonso U. Rubrico Project Leader: Dr. Franz A. De Leon
Implementing Agency: Advanced Science and Technology Implementing Agency: Advanced Science and
Institute Technology Institute
COMPLETED
AI for Health and Education
Development of a Multi-lingual Chatbot for Health Enhancing Man-Machine Interaction through
Monitoring of Public School Children Conversational Agents

Project Leader: Dr. Judith J. Azcarraga Project Leader: Ms. Ethel Chua Joy Ong
Implementing Agency: De La Salle University Implementing Agency: De La Salle University
Status: Completed Status: Completed

Validation Process with Children

COMPLETED

AI for Education
FilWordNEt: Diachronic Representation and Project iWag: A Mobile Web Bidirectional
Linguistic Study of Filipino Word Senses Neural Machine Translation System for Filipino
Across Social and Digital Media Contexts and Cebuano

Project Leader: Dr. Briane Samson Project Leader: Ms. Kristine Mae Adlaon
Implementing Agency: De La Salle Implementing Agency: University of
University (DLSU) Immaculate Concepcion
COMPLETED

AI and AI Robotics for Industries

Intelligent Structural Health Monitoring Development of a CNN and RNN Topology

via Mesh of Tremor Sensors (MESHM) for Impedance Spectroscopy Analysis

Project Leader: Dr. Marnel Peradilla Project Leader: Engr. John Paolo Ramoso
Implementing Agency: De La Salle Araneta University Implementing Agency: University of the
Philippines Los Baños
COMPLETED

AI and AI Robotics for Industries

Autonomous Societally Inspired Mission Oriented Vehicles (ASIMOV)

Project 1: Harmonized Aerial Watch and Project 2: Robot for Optimized and Autonomous
Knowledge-based Survey (HAWKS) Mission-Enhancement Responses (ROAMER)
Project Leader: Dr. Jose Ildefonso Rubrico Project Leader: Engr. Peter Banzon
Implementing Agency: DOST Advanced Science Implementing Agency: DOST Advanced Science and
and Technology Institute Technology Institute

AI-based Computer Vision

COMPLETED
AI for Mobility

Design and Development of Intelligent A Lightweight Edge Computer Vision Solution

Traffic Control and Management System for Smart and Efficient Traffic Management
Project Leader: Dr. Junrie B. Matias Project Leader: Engr. Alex Llave Maureal
Implementing Agency: Caraga State University Implementing Agency: University of Science and
Technology of Southern Philippines –Cagayan de Oro
COMPLETED

AI for Disaster
Risk Reduction
Development of an AI-assisted
Thunderstorm Early Warning System
from Analysis of Doppler Radar Data

Project Leader: Dr. Prospero Naval, Jr.

Implementing Agency: University of the Philippines
Diliman
ONGOING

AI for Disaster
Risk Reduction
AI-4RP: AI-powered weather
forecasting for Resilient Philippines

Project Leader: Ms. Cristina Uson

Implementing Agency: DOST-PAGASA
Co-implementing Agency: DOST-ASTI
ON-GOING

AI for Disaster Risk Reduction

Accelerated Earthquake Multi-hazards Mapping and Risk Assessment Program of the Philippines (ACER) Program

Deterministic Liquefaction and Tsunami Hazard

Active Fault Mapping Using Integrated Geophysical,
Mapping of Selected Low-lying Areas in the Philippines
Geomorphological, and Paleoseismological
using GIS, AI, and Geophysical Techniques
Methods
Project Leader: Maria Isabel T. Abigania
Project Leader: Kathleen L. Papiona
Implementing Agency: DOST-PHIVOLCS
Implementing Agency: DOST-PHIVOLCS
ON-GOING

AI for Disaster Risk Reduction

Accelerated Earthquake Multi-hazards Mapping and Risk Assessment Program of the Philippines (ACER) Program

Deterministic Ground Shaking and Earthquake-Induced

Spatio-temporal Land Cover Mapping for Risk Landslide Hazard Modeling Using Geophysics, Remote
Assessments in Land Use Planning through AI (LUPA Sensing and Artificial Intelligence
Project)
Project Leader: John Harold Bajador Tabuzo Program and Project Leader: Dr. Arturo S. Daag
Implementing Agency: DOST-PHIVOLCS Implementing Agency: DOST-PHIVOLCS
 NEW

Digital Reinforcement to Enable AI Nexus for Advanced Risk

(Artificial Intelligence) to Maximize Reduction and Analytics System
Solutions in BJMP Region X for PNP-COCPO

Project Leader: Prof. Nathalie Joy G. Casildo Project Leader: Engr. Jodie Rey D. Fernandez

Implementing Agency: Central Mindanao University Implementing Agency: Philippine National Police –
Cagayan de Oro City Police Office (PNP-COCPO)

Development of a Meranaw COULIGLIG: Cooperative Unified

Speech Corpus Logistics using Intelligent Grouped
Robots

Project Leader: Engr. Karla Morales Madrid-Khalid

Project Leader: Engr. Rudolph Joshua U. Candare
Implementing Agency: Mindanao State University - Marawi
Implementing Agency: Caraga State University
 AI for Smart City
A Link-Up of Geomatics and Social MINERVA: Monitoring of INdicators PATURO: Platform for Assessment
Science Research for the Development for Efficient Redevelopment and and Tracking of Urbanization –
of Smart Cities (LUNGSOD) Value Assessment Related Opportunities

Project Leader: Czar Jakiri S. Sarmiento, Ph.D.

Implementing Agency: Training Center for Applied Project Leader: Prof. Christopher Monterola, Ph.D. Project Leader: Erika Fille T. Legara, Ph.D.
Geodesy and Photogrammetry (TCAGP) - University of Implementing Agency: Asian Institute of Implementing Agency: Asian Institute of Management
the Philippines Diliman Management Cooperating Agencies:
Cooperating Agencies: Cooperating Agencies: City Government of Cauayan, Isabela and Isabela State
Local Government of Iloilo City City Government of Baguio City, and University of University – Cauayan Campus
Department of Science and Technology - Western the Philippines Baguio Campus
Visayas
University of the Philippines Visayas
 Good Governance through Data Science and Decision Support System
GODDESS Program

20 Completed
Projects PhP1.0M

A decision-support system on
preemptive evacuation plan for flood
PhP997,500

A smart garbage collection and

monitoring system
A platform for the social
City in Region 8 competitiveness

PHP35,764,699.40 disaster integrating Covid-19 zoning

amelioration datasets of
beneficiaries of 4Ps
Total Investments DOST SETUP analytics
PhP999,982.40 PhP1.0M

An employment and services portal A national park management system A Data-Driven Student Support
PhP999,872.00 System using Admission to
Graduation Data System that can predict the
A data analytics system for PhP4.9 nutritional status among under five
Philippine Rice Genetic Sources PhP1.0M M children

An operational Data Analytics A data-driven decision-making in

An automated tariff collection system
system for PhilRice improving school effectiveness
PhP1.0M based on student retention A Queen Pineapple Data Analytics &
metrics Recommender System
An automated process for monitoring PhP999,800
persons with disabilities (PWDs) PhP4.9M

An information system on the A vision-based traffic monitoring An integrated Barangay

2 scholarship services monitoring
social welfare services for system for traffic administrators Information Management System
senior citizens
AI Based Startups and Spin-Offs
 CALL FOR Wealth Creation: Industry Solutions and
PROPOSALS 2025 Competitiveness Development

Artificial Intelligence
a.AI for Banking and Finance
• AI-driven financial inclusion
• Fraud detection and prevention
• Microfinance optimization
• Regulatory technology
• Convergence of AI and blockchain for secure and transparent financial systems
• Financial risk prediction and management
b.AI in Cybersecurity
• AI-driven Threat Detection and Prevention
• Cybersecurity in Critical Infrastructure using AI
• AI for Phishing Detection and Prevention
• AI for Cybersecurity in IoT Ecosystems
• Privacy-Preserving AI in Cybersecurity
c.Beyond Words: Advancing Large Language Models
• Corpus and LLM Platform for KWF-Targeted Languages to build datasets and models for Philippine
languages.
• Enhancing LLMs’ Abilities for Underrepresented or Endangered Languages to preserve cultural
heritage and linguistic diversity.
• Building Stronger Cross-Lingual and Multilingual Capabilities to improve LLMs’ ability to operate in
diverse linguistic environments.
• Applications of LLMs in Specialized Domains to develop solutions for education, healthcare, and other
sectors.
 Prospects for Partnership
S&T Partnership

Collaborative Policy development Co-Mentoring in AI Other collaborative

Research for growing AI programs/ Projects by activities among
and Development Experts and Industry those working in AI
(R&D) in AI Stakeholders

51
DOST Support for the Artificial Intelligence Industry
Research and Development (R&D) and Innovation Support
Support for Research and Development

DOST and PCIEERD

Grants-in-Aid Program
2025 Call for Proposals

Support for Technology Transfer and Support for Human Resource and
Commercialization Institution Development
With the right policies,
infrastructure, and partnerships,
the government can unlock AI's
full potential to improve people's
lives and advance the country's
sustainable development goals.
Content
1. AI in the Philippines & the ASEAN
• Ethical AI Development and Governance
• OneDOST4U! Providing Solutions, Opening Opportunities for
the AI Sector through STI
2. DOST-ASTI and its AI Initiatives
•AI projects
•Demos
 DOST-ASTI has been at
the forefront of local
R&D initiatives as a
technology enabler.

Our Mandate
The Advance Science and
Technology Institute (ASTI) is a
research and development
institute of the Department of Conduct R&D work Complement the
Undertake long-
Science and Technology term research to in advanced fields overall endeavor in
mandated to undertake R&D strengthen and of studies the scientific field
activities aimed at strengthening modernize the with intensive
and modernizing science and Philippines’ science activities in
technology infrastructure in the and technology computer and
advanced fields of Information infrastructure information
technologies
and Communication Technology
and Electronics.
 Research and Innovation for
National Development
SPACE TECHNOLOGY WIRELESS TECHNOLOGY ARTIFICIAL INTELLIGENCE

Remote Sensing Synthetic Aperture Radar and ASTI Advanced Remote Resilient Education Information Signal Assessment Using Understanding Lightning ASTI-Automated
and Data Science Automatic Identification System for
Innovative Terrestrial Monitoring and Data Acquisition Infrastructure for the New Normal Geospatial Analysis and Thunderstorms Labeling Machine iTANONG Gul.ai
(DATOS) Helpdesk Maritime Surveillance (SARwAIS) (ASTI-ARQ) (REIINN) Project (SAGAP) Project (ULAT) (ASTI-ALaM)

EMERGING TECHNOLOGIES ROBOTICS SMART CITIES

Autonomous and Robot for Optimized and

DataProtect Quantum Circuit Kooha NetMesh Autonomous Mission- V2X Initiatives for Road
Unmanned Safety (VIROS)
Simulation (QCS) Systems (AUS)
Enhancement Responses
(ROAMER)
S&T Infrastructure for Research and Innovation
National Development

Climate Disaster Risk Agricultural Urban

Change Reduction & Production Planning
Assessment Management Optimization

Blockchain Smart & Quantum

Artificial Internet of Robotics &
Autonomous Systems
Sensor Networks &
Early Warning Systems Technology
Big Data Sustainable Cities Computing
Intelligence Things

Science & Technology Infrastructure

Computing and Archiving Philippine Research, Education Electronics Product Philippine Earth Data PhilSensors and Meteorological Data
Research Environment and Government Information Development Center Resource Observation Acquisition Stations for Information
Dissemination (MASID)
(COARE) Network (PREGINET) (EPDC) Center (PEDRO)
 Artificial Intelligence is EXPENSIVE

S&T Infrastructure Technical Expertise & Model Development and Regulatory and
Talent Acquisition Lifecycle Compliance

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

ACCORD CONNECT
AI Center for Computation, Optimization, and Collaborative Open Network Infrastructure for
Responsible Research Development Emerging Technologies, Computing, and
Connectivity

ASCENT
Advancement of Autonomous Systems, STRATEGIC THRIVE
Wireless Connectivity, and Embedded PROGRAMS Translating Research into Valuable and
Electronics for National Transformation Helpful Everyday Applications
IMPACT-
SIMULA
RP-TROPA
www.asti.dost.gov.ph
...moving beyond possibilities
 DOST-ASTI

Decentralized Intelligent Model

Exchange Repository (DIMER)

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

Gul.AI
Gul.AI Project: AI- and IoT-Assisted Small-scale Plant-growing System

The Gul.AI Project is a small-scale plant-growing system for conducting research experiments on
small crops.

The setup showcases the application of the following technologies to plant-growing:

• Internet of Things (IoT) for data gathering through remote and automated operation of the
system
• Artificial Intelligence (AI) and statistical analytical methods to the analysis of multivariate
datasets gathered by the system.
Through experimentation using the system, researchers will be able to identify ideal plant-
growing conditions for specific variants of small crops.

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

SAqFeeder
Smart Aquafarm Feeder and Monitoring System in Highly Turbid Conditions
SAqFeeder aims to optimize the feeding process by developing an intelligent feeding system using computer
vision, auditory filtering, and automated actuation, addressing the challenges in water turbidity.

It addresses the need for an improved monitoring

system in aquafarms particularly in turbid conditions
where challenges include forward and backward
scattering, absorption, large particles, and artificial
lighting which hinders effective monitoring,
particularly in assessing feed quantities, which
impacts the status of aquafarms, likewise the
economic sustainability of farmers.
...moving beyond possibilities www.asti.dost.gov.ph
 DOST-ASTI

AI-4RP
AI-Powered Weather Forecasting for a Resilient Philippines

A project co-implemented by DOST-ASTI and DOST-PAGASA aimed at

enhancing weather forecasting in the Philippines using advanced AI
technologies, complementing the current and traditional weather
forecasting methods of the Philippines, with the following key features
that are expected to assist in improving early warning systems and
better managing the impacts of meteorological hazards and disasters:
• High Resolution: Improving forecast accuracy by increasing
horizontal resolution from 3 km to 2 km
• Extended Lead Time: Extending the forecast lead time from 2 days
to 14 days
• Faster Processing: Reducing forecast runtime from 3 hours to 15
minutes

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

MaSense Challenge Solution

Intelligent Machine Condition Monitoring System
for Limestone Grinding Industries in the Philippines, including The MaSense project develops an
small and medium enterprises (SMEs), intelligent machine monitoring system and
face high costs and limited access to a visualization dashboard tailored for local

advanced machine condition settings, while advancing machine

analytics curricula in higher education to
monitoring tools and training, resulting
cultivate a workforce capable of
in inefficiencies, costly downtime, and a
implementing predictive maintenance
lack of skilled professionals in predictive
technologies.
maintenance.

Objectives Impact
• Conduct research and development on MaSense enhances operational
an intelligent machine condition efficiency for local industries, reduces
monitoring system.
machine downtime, and builds a
• Develop a dashboard monitoring
sustainable pipeline of skilled
platform as a testbed for integrating
professionals, empowering the
and testing various sensors.
• Create a curriculum on advanced Philippines to adopt modern, cost-
machine predictive analytics for higher effective predictive analytics
education and technical-vocational technologies for industrial growth.
institutions.

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

Cerv.ai
Cerv.ai: Identification of Regions-of-Interest in
Cervical Pre-Cancer and Cancer Screening Images Challenge Solution
Using Computer Vision
Cervical cancer remains a By leveraging computer vision/deep
preventable yet significant health learning models and expertly
concern in the Philippines, with low annotated datasets, Cerv.ai
screening rates and time-consuming enhances diagnostic accuracy,
diagnostic processes delaying early reduces screening time, and
detection and impacting patient improves patient comfort, making
outcomes. early detection more efficient and
accessible.

Objectives Impact

• Train Medical Professionals in Cerv.ai empowers medical

Annotation professionals and healthcare providers

• Create an Annotated Image with AI tools that streamline pre-cervical

cancer screenings, improving workflow
Dataset
efficiency and diagnostic accuracy.
• Develop and Evaluate AI Patients benefit from faster, more
Models reliable screenings, leading to earlier
• Document Project Findings interventions and better treatment
outcomes.

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

ROAMER
Robot for Optimized and Autonomous Mission-
Enhancement Responses

The ROAMER mobile robot supports banana

plantations in monitoring and detecting
banana plant diseases by using data-driven
decision making based on computer vision,
sensing, and navigation, therefore, ensuring
crop health and improving productivity.
ROAMER also aids farmers in plant care and
disease management by utilizing a robot arm
that uses a precision spraying pipeline, thus
reducing the exposure of humans to chemical
hazards and uses farm chemicals efficiently.

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

VIROS - STEER
V2X Initiatives for Road Safety -
Smart Traffic Electronic Enforcer on the
Roads
The VIROS Project integrates AI
and emerging tech to transform
traffic control systems into data-
driven models. Key technologies
include AI-powered Computer
Vision and Vehicle-to-Everything
(V2X), enabling seamless
communication between vehicles
and road users, improving road
safety for everyone.

www.asti.dost.gov.ph
...moving beyond possibilities
 DOST-ASTI

Leveraging Advanced NLP and Modern

Information Retrieval Techniques for
Organizational Data Efficiency for Filipinos

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

A Natural Language Interface

to Databases for Filipinos
iTANONG is an AI web application that answers questions in Tagalog,
English, or Taglish using information from organization-maintained
relational databases and uploaded documents.

It aims to build a natural language querying engine and

linguistic interface, conduct AI and data science research,
facilitate capacity building, promote technology adoption,
and commercialize the generated products and technologies.

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

ITANONG Briefer

Most business Many applications are Allow non-technical Make searching for
data is relational. built on relational end-users to query answers easier by
databases. information relevant using language more
to their functions in an comfortable to the
organization. users.

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

How it works

Natural Language
Question

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

Existing issues in real-life settings

A large amount of Multiple modes of Linguistic diversity in Mobility, ease of

organizational data communication the Philippines access, and inclusivity
still exists as flat files

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

SkAI-Pinas
Philippine Sky Artificial Intelligence Program ASTI-ALaM

ASTI-Automated Labeling Machine (ASTI-ALaM)

aims to democratize access to cutting-edge AI
technologies to address critical challenges in AI
adoption in the Philippines—such as the need for
technical expertise and costly hardware to develop
robust AI models.

ALaM-Large Scale Initiative (ALaM-LSI) adopted AI-

driven solutions for the automated labeling and
segmentation of remote sensing data. Using pre-
trained AI models, ALaM-LSI accelerates the
process of preparing high-quality labeled datasets,
Democratized Intelligent
enabling more efficient data utilization for
Model Exchange Repository
geospatial applications.
(DIMER)

...moving beyond possibilities AI-based Processing Interface (AIPI) www.asti.dost.gov.ph

 DIMER

...moving beyond possibilities www.asti.dost.gov.ph

 Intelligent Flood Detection Problem Solution
using Traffic Surveillance • Detect and classify flood • Three-class classification: Normal,

Images based on conditions in near real time using Raining, Flooding.

• Preprocessed with DeepLabv3
Convolutional Neural existing traffic CCTV infrastructure.
• Leverage computer vision to (semantic segmentation), then

Network and Image Parsing automate flood detection for classified using a custom CNN (4
conv layers + 2 FC layers).
quicker, more accurate local
• Distinguish flood vs. no flood while
government response.
accounting for rain presence.
Eduardo Jr Piedad; Elmer Peramo; Jeffrey Aborot; Joshua Russel
Bensig; Paulyn Jamila Deiparine; Stephanie Marie Flores

Model Dataset

• Cascaded approach: DeepLabv3 • 30,000 images (10,000 per class)

(frozen) + newly trained CNN on from Mandaue City CCTV
segmented images. (day/night, normal/raining/flood).
• Trained 10 CNN models, selecting • 80% for training/validation, 20%
best performer. testing.
• Best model metrics: ~80.67% • Annotated with DeepLabv3 for
accuracy, 86.33% precision, 81% object segmentation.
Scene parsing of raw daytime CCTV images (left); night-time CCTV
recall, 79.67% F1-score.
images (right).

...moving beyond possibilities www.asti.dost.gov.ph

https://doi.org/10.1109/ICOCO56118.2022.10031718
 A deep hybrid GNN based on Problem Solution
edge-conditioned and graph • Prostate cancer at metastatic • Binary classification (active vs.

isomorphism network stages has low survival rates; new inactive) of small molecules using
graph-based data.
convolutions for PC-3 drugs must target tumor cells
while sparing normal cells. • Deep Hybrid Graph Neural Network

anticancer screening • Leverage deep learning to (DHGNN) with GIN + ECC layers,
global pooling, and FC layers.
accelerate anticancer drug
Adrian S. Remigio; Jeffrey Aborot; Jonel Hong • Optimized via Adam, aiming for
discovery for PC-3 tumor
precise identification of active
treatment.
compounds.

Model Dataset

• DHGNN: 3 convolutional layers • 6000 molecular graphs (1568

each in GIN and ECC branches active, 4432 inactive) from NCI’s
(50 embedding channels/layer), PubChem library.
plus 3 FC layers. • Node/edge features from Yan et
• Performance: Accuracy = 0.883, al. (2008), accessed through the
AUC = 0.828, Precision = 0.808, Spektral Python package.
Recall = 0.713.
Schematic diagram illustrating the prediction workflow and network
architecture of the Deep Hybrid GNN model.

...moving beyond possibilities www.asti.dost.gov.ph

https://doi.org/10.1109/SPIN57001.2023.10116850
 Multiple Pedestrian Object Problem Solution
Recognition to Correct User • Visually impaired individuals need • Multi-class object detection for

Direction of Vision-Impaired reliable orientation at pedestrian pedestrian-related objects (lights,

signs, lanes).
Assistive Device crossings.
• This system detects lights, signs,
• Geometric post-processing checks if
detected objects align correctly in the
and lanes to align the user’s
camera frame.
facing direction, ensuring correct • Implements YOLOv8 (CSPDarknet +
Eduardo Jr Piedad; Samuel Harrison P. Cerrudo; Elmer C. Peramo orientation to cross safely. PAFPN) and Faster R-CNN (ResNet50 +
FPN), both pre-trained on MS COCO.

Model Dataset

• Two object detection approaches: • Curated images from

YOLOv8 vs. Faster R-CNN. public/smartphone sources.
• YOLOv8: Precision 88.8%, Recall • Annotations for Green/Red
90.8%, mAP@0.5 = 80%. Pedestrian Lights, No Pedestrian
• Faster R-CNN: Precision 83.6%, Crossing Signs, Pedestrian
Recall 86.2%, mAP@0.5 = 78.8%. Crossing Signs, and Lanes (160+
instances each).

Fine-tuned YOLOv8 bounding boxes detections on sample images.

...moving beyond possibilities www.asti.dost.gov.ph

https://doi.org/10.1109/ICOCO59262.2023.10397854
 Combining First-Order and Problem Solution
Texture Features with • Classify aerial images of • DHFFNN (Deep-Handcrafted
Convolutional Neural residential areas (dense, medium, Feature Fusion Neural Network):

Network for Classification of sparse) based on structure

concentration.
Parallel branches for ResNet-50
(deep features) and MLP (273D
Residential Area Density in • Investigate whether combining handcrafted features).

Aerial Images CNNs with handcrafted features

(FOS, GLCM, GLRLM, GLSZM, GLDM,
• Outputs are concatenated into a
final classification layer for
NGTDM) can boost accuracy. residential density.
Adrian S. Remigio; Jonel Hong; Jeffrey Aborot

Model Dataset
• Two-branch design: • NWPU-RESISC45: 31,500 images
⚬ CNN branch: ResNet-50 extracting deep
(256×256 px).
features (global average pooling).
⚬ MLP branch: Processes handcrafted • AID: 870 images (600×600 px,
features. rescaled to 256×256).
• Final classification outperforms baseline CNN
and classical ML with handcrafted features:
⚬ NWPU-RESISC45: ~99.6% accuracy (vs.
Schematic overview of the methodology implemented in this study. The input ~91.2% CNN-only).
aerial images are mapped into a vector representation of fos and texture features, ⚬ AID: ~98.1% accuracy (vs. ~97.3% CNN-only).
followed by an application of cml models to predict the output label.

...moving beyond possibilities www.asti.dost.gov.ph

https://doi.org/10.1109/CSDE59766.2023.10487144
 Optimizing Healthcare Problem Solution
Accessibility: An Integration • Improve hospital accessibility in Olongapo • Land-use Segmentation: 23 classes (e.g.,

of GIS and Machine Learning City by locating underserved areas and

potential new facility sites.
buildings, vegetation, water) using U-Net on
the Semantic Drone Dataset.

for Strategic Hospital Facility ⚬ Shortest-route analysis from population

centers to existing hospitals.
• U-Net Variants: ResNet-34, ResNeXt-50, VGG16
encoders, each tested with/without data

Location Planning ⚬ Determining coverage (buffer zones) to

identify poorly served regions.
augmentation and a learning rate scheduler.
• GIS Methods: QGIS shapefile
boundary/overlay, QNEAT3 for route analysis,
⚬ Semantic segmentation of land use via
5–10 km buffer coverage analysis for
drone imagery to find suitable terrain hospitals.
Kahlil Sebastian G. Ramos; Elmer C. Peramo for new hospitals.

Model Dataset

• Trained Models: 6 U-Net models (3 • Sources:

encoders × 2 augmentation ⚬ Healthcare Facilities: DOH - NHFR
settings). (Zambales)

• Best Performer: ResNet-34 U-Net (no ⚬ Population Density: Humanitarian

Data Exchange
augmentation)
⚬ Road Network: OpenStreetMap
⚬ Training Loss: 18.37%, Accuracy:
(OSM)
94.16%
⚬ Land Use: Semantic Drone Dataset
⚬ Validation Loss: 91.14%, Accuracy:
Model inference for ResNet-34 U-Net model without data augmentation (400 images, 23 classes)
79.17%
using a satellite image of Olongapo City, Zambales.

...moving beyond possibilities www.asti.dost.gov.ph

https://doi.org/10.1109/ICAIIC60209.2024.10463219
 Simpler Machine Learning Problem Solution
Methods Outperform Deep • Detect multiple induction motor faults • Five-class classification (1 healthy + 4 fault

Learning in Motor Fault using current signals under varying

load (0–100%).
types) across five load levels.
• CNN Approaches: Convert signals to 2D

Detection • Compare classical ML vs. CNN-based

approaches, as prior CNN methods
Frequency Occurrence Plots (FOP) or
Recurrence Plots (RP).
• Traditional ML Approaches: Train on 1D
achieved only ~80% accuracy.
Antonio C. Briza; Eduardo Piedad; Elmer C. Peramo Fourier-transformed signals (Logistic
• Investigate if simpler ML models can
Regression, KNN, Naive Bayes, Decision
outperform CNNs in fault classification.
Tree, Random Forest, SVM, MLP, Extra Trees,
LightGBM, AdaBoost).

Model Dataset

• Best Model: LightGBM with 93.20% • 5 Motor Conditions (healthy +

accuracy, outperforming CNN- misalignment, short circuit, broken rotor,
outer bearing).
based methods (~74.80–82.80%).
• 5 Load Levels (0%, 25%, 50%, 75%, 100%),
• 10 classical ML models + 1 new CNN-
totaling 3,750 samples (150 per condition ×
RP approach vs. 2 previous CNN-FOP 5 loads).
Illustration of the different preprocessed datasets used by the methods models. • Preprocessing: 90% percentile clipping, log
in this study. The images represent one sample of a healthy motor with • LightGBM also leads in F1-score, normalization, Discrete Fourier Transform
a loading capacity of 75%.
precision, and recall for each fault (DFT); CNN input requires 2D plots (FOP or
class. RP).

...moving beyond possibilities www.asti.dost.gov.ph

https://doi.org/10.1109/ICAIIC60209.2024.10463445
High-Resolution Landslide Problem Solution
Mapping Using Bi-Temporal • Many mountainous regions face • Task: Change detection and semantic

Imagery frequent landslides, and manual

detection is slow and incomplete.
segmentation to identify landslide-
affected areas.

• Delayed response increases risks for • Architecture: Bi-Temporal Adapter

Samuel Harrison P. Cerrudo Network (BAN) with frozen base models
communities.
(e.g., CLIP), specialized adapter
• Use Bi-Temporal Adapter Network on
branches, and bridging modules to
pre-/post-event imagery for faster,
handle time-separated imagery.
more accurate landslide detection.

Model Dataset

• Trained Model: One primary BAN • Name: Global Very-High-Resolution

model. Landslide Mapping (GVLM) Dataset.
• Performance: Precision = 0.88, • Provider: Researchers led by Dr.
Recall = 0.85, F1-score = 0.86. Xiaokang Zhang.

• Parameters: ~34 million. • Scope: 17 bi-temporal image pairs at

0.59 m resolution, each capturing
pre- and post-landslide events.
Kantagnos Landslide (April 11, 2022): Bi-Temporal AI analysis
pinpoints landslide-affected zones for proactive disaster response.

...moving beyond possibilities www.asti.dost.gov.ph

Automated Building Problem Solution
Footprint Segmentation for • Many regions lack current land cover • Task: Semantic segmentation of

Disaster Resilience maps, hindering environmental

monitoring, resource management, and
building footprints from aerial imagery.
• Architecture: U-Net with EfficientNet-B1
Samuel Harrison P. Cerrudo urban planning. backbone, trained using Focal Loss +
• Mask2Former-based model for multi- Dice Loss to address class imbalance.
class segmentation, aiming to improve • Augmentation: Rotations, flips, and
land use classification and resource color jittering to boost model
planning. robustness.

Model Dataset

• Trained Model: One primary U-Net. • Name: Open Cities AI Challenge

• Performance: ~0.86 Jaccard Index Dataset (GFDRR).
(IoU) on the test set. • Scope: High-resolution drone
• Parameters: ~8 million imagery from 12 African

(EfficientNet-B1 backbone). cities/regions.

• Coverage: Over 700,000 buildings
labeled for segmentation.

Automated U-Net segmentation highlights building footprints for

improved disaster risk planning and urban management.

...moving beyond possibilities www.asti.dost.gov.ph

Comprehensive Land Use Problem Solution
Land Cover Classification • Disaster-prone areas need precise • Task: Multi-class semantic segmentation
building footprint data for planning and of land cover types in satellite imagery.
Samuel Harrison P. Cerrudo emergency response, but manual • Architecture: Mask2Former with a
mapping is slow and prone to errors. ResNet-50 backbone, leveraging multi-
• Training a U-Net on high-resolution scale attention for localized texture
imagery enables large-scale, accurate detail.
building footprint segmentation, • Approach: Large-scale satellite data
enhancing disaster resilience and training to better distinguish complex
urban development. land cover categories.

Model Dataset

• Trained Model: One primary • Name: OpenEarthMap

Mask2Former model. (OpenEarthMap Consortium).
• Performance: Mean Intersection • Scope: Over 50,000 labeled
over Union (mIoU) = 83.2% across polygons spanning multiple land
12 classes. cover classes.
• Parameters: ~29 million
(Mask2Former + ResNet-50

Mask2Former-generated land cover segmentation reveals diverse

backbone).
classes for enhanced environmental monitoring and urban planning.

...moving beyond possibilities www.asti.dost.gov.ph

Rice Field Monitoring and Problem Solution
Mapping • Frequent cloud cover and sparse • Task: Crop-type classification and semantic
segmentation to distinguish rice fields from
ground data make it hard to
Gary Chris D. Lacdang other land covers.
accurately map and monitor rice
• Methods:
fields in near-real time. ⚬ Deep Attentive Time Warping (DATW): Uses
• Use multi-temporal Sentinel satellite attention to extract key time-series

imagery and deep learning to segments, creating pseudo-labels.

⚬ ConvLSTM: Combines CNN spatial features
produce precise rice field maps
with LSTM temporal dynamics, refining
under varied weather conditions. DATW outputs into full-image rice field
maps.

DATW ConvLSTM-SAR

Model Dataset
• Trained Models: • Name & Provider: Combined Sentinel-1
⚬ DATW for pseudo-label generation. (SAR) & Sentinel-2 (Optical) data from ESA,
⚬ ConvLSTM-SAR and ConvLSTM-Optical for final
with PRiSM field data for ground truth.
classification.
• Time Frames:
• Performance Metrics:
⚬ DATW: F1 = 94%, Precision = 99%, Recall = 91%, ⚬ Sentinel-1: Aug 2017–Mar 2018
Accuracy = 98%. ⚬ Sentinel-2: Aug 2019–Mar 2020
ConvLSTM-Optical Final Output ConvLSTM-SAR
⚬ ConvLSTM-SAR: F1 = 94%, Precision = 91%, IOU = • Processing: DATW pseudo-labels and final
90%, Recall = 98%, Accuracy = 96%. labeled sets for ConvLSTM training/testing.
⚬ ConvLSTM-Optical: F1 = 84%, Precision = 82%, IOU
DATW isolates key time-series features, ConvLSTM-SAR leverages radar for cloud-free = 73%, Recall = 86%, Accuracy = 86%.
mapping, and ConvLSTM-Optical provides high-fidelity segmentation—together
enabling precise rice field monitoring in Region 3 Central Luzon.

...moving beyond possibilities www.asti.dost.gov.ph

Exploring a Framework for Problem Solution
Vehicle Counting and • Rapid urbanization in the Philippines • Task: Multi-class vehicle detection from

Detection in the Philippine leads to heavy traffic congestion,

and manual vehicle counting is
CCTV frames.
• Approach: YOLOv11 (a next-gen YOLO

Setting Using Data Processing labor-intensive and error-prone.

variant) fine-tuned on a custom-labeled
Philippine vehicle dataset.
and Data Augmentation • Develop an AI-driven vehicle
detection/counting system tailored
• Data Augmentation: Grayscale

Techniques
transformations, blocking non-target
to unique local vehicle types (e.g., vehicles to increase jeepney instances, and
jeepneys, tricycles). class grouping for similar vehicle types.
Ethyl Joy I. Badiang; Antonio C. Briza; Elmer C. Peramo

Model Dataset
• Trained Model: YOLOv11 pre-trained on • Sources: CCTV footage from MMDA
COCO, then fine-tuned on local Philippine (Metro Manila, 6 AM–8 PM, two days),
data.
and additional recordings in Quezon
• Performance:
City.
⚬ Baseline Precision: ~15–21% (day/night).
⚬ Grayscale (Day): 52% Precision.
• Classes: 19 vehicle types (bicycle,
⚬ Class Grouping (Day): 67.53% Precision, bus, car, jeepney, tricycle, etc.).
61.76% Recall (best day result); (Night): • Annotations: Done via CVAT; includes
54.5% Precision, 47.1% Recall. open-sourced data with 9 core
MMDA traffic surveyors carry out manual vehicle counts, highlighting
the need for AI-driven automation in Philippine roads.
classes.

...moving beyond possibilities www.asti.dost.gov.ph

Vehicle Route Optimization Problem Solution
Using Heuristics and Deep • Manual assignment of daily vehicle routes • Task: Vehicle route optimization under

Learning leads to inefficient travel distances and

resource use.
capacity constraints to minimize total
travel distance.
• Multiple deliveries under capacity constraints • Approach: The Transformer model
become complex as fleets grow.
generates partial routing solutions, then
• A deep-learning “Learning to Delegate”
heuristic solvers (LKH, HGS, OR-Tools) refine
Antonio C. Briza Transformer combined with heuristic solvers
these routes in a deep-learning–heuristic
(LKH, HGS, OR-Tools) reduces route distances
pipeline.
and improves operational efficiency.

Model Dataset

• Trained Models: Three Transformer • Name: Local Logistics Daily

variants integrated with subsolvers. Routing Assignment.
• Performance: • Provider: A local logistics
⚬ Transformer + LKH: 48% reduction company.
in manual route distance. • Scope: 154 days (June–November
⚬ OR-Tools Heuristic: 59%
2022) of daily vehicle assignment
reduction.
CSV files.

Visualization of direct straight-line route assignment (top) and

OSMnx-based shortest path (bottom left).

...moving beyond possibilities www.asti.dost.gov.ph

Digital Preservation of Insect Problem Solution
Specimens via 3D • Physical insect specimens • Task: 3D reconstruction of insect

Reconstruction deteriorate over time, limiting specimens using self-supervised

learning approaches.
research and educational use.
Jose Maria E. Braga • Creating digital 3D models • Architecture and Methods Used:

preserves fine details and ⚬ 3D Gaussian Splatting: A

rasterization-based technique
increases accessibility for
optimizing Gaussian kernels in 3D
students, researchers, and the
space to reconstruct surfaces.
public.

Model Dataset

• Approach: A 3D Gaussian • Name: UPLB Museum of Natural

Splatting pipeline. History Insect Specimens.
• Metrics: PSNR (Peak Signal-to- • Provider: University of the
Noise Ratio) and SSIM (Structural Philippines Los Baños Museum of
Similarity Index Measure). Natural History.
• Scope: Multi-angle images of
insect specimens for self-
High-resolution, multi-angle photos captured at the UPLB Museum of supervised 3D modeling.
Natural History form the basis for 3D insect reconstruction.

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

Project Overview
The DOST Nexus for Artificial
Intelligence Research and Existing S&T
Infrastructure
Applications (DOST-NAIRA) is designed Provide
Consultancy
to establish a comprehensive AI virtual Services

hub in the Philippines. The primary aim

is to harmonize, integrate, and build
upon existing AI initiatives such as
COARE, PREGINET, ALaM, and iTanong, DOST-NAIRA
AI Virtual Hub
enhancing collaborative AI research and
development (R&D), providing
consultancy services, and attracting
Enhance
global R&D firms to the country. Collaborative
Attract Global
R&D Firms
AI R&D

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

Project Objectives

Build an AI Strengthen the Build AI Enhance Public Promote Create a

Virtual Hub AI Ecosystem Capacity Services and Ethical AI Sustainable AI
To create an integrated To establish a To develop a skilled Productivity Practices Ecosystem
virtual hub that will centralized AI hub that workforce in AI and To improve the To ensure responsible To establish a self-
serve as a fosters collaboration related technologies efficiency and AI use by promoting sustaining AI
collaborative platform among universities, through training effectiveness of public the development and ecosystem through
for AI solutions industries, and programs, workshops. services through AI- deployment of AI technology licensing,
workflow. government agencies. driven solutions. technologies in ways commercialization of
that are transparent, AI products, and
and fair. collaborations.

...moving beyond possibilities www.asti.dost.gov.ph

 DOST-ASTI

Technology Roadmap
Methodology/Workplan
2029 (Maturity and
Sustainability)
2027-2028 (Expansion and • AI Ecosystem Maturation:
Scaling) • Establish a self-sustaining AI
ecosystem through technology
• Enhanced AI Services: licensing, commercialization of AI
2025-2026 (Initial Development • Scale the AIaaS platform, expanding products, and innovation.
and Implementations) model and dataset repositories to • AI-Driven Innovations:
• AI Virtual Hub Setup: support broader sectors • Incorporate advanced technologies
• Establish the AI virtual hub • Focus on improving the efficiency of like Generative AI and Large
• Develop AI as a Service (AIaaS) platform public services using AI solutions. Language Models (LLMs) to further
• Target: TRL 5-7 (Technology ready for testing • Commercialization & Collaboration: enhance the platform's capabilities.
and demonstration). • Begin commercialization of AI tools • Sustained Capacity Building:
• Capacity Building: and technologies developed • Continuously update training
• Train at least 600 students, professionals, and through the hub.
solution developers on AI.
programs to keep the workforce at
• AI Research Hub:
• Forge partnerships with local the cutting edge of AI technology.
• Foster collaborations between universities,
industries and expand • Ensure continued focus on
industries, and government agencies to collaborations. empowering disadvantaged
promote interdisciplinary AI research. • Target: TRL 8 (Demonstration in communities through accessible AI
• Begin feasibility studies and pilot research operational environment). tools.
projects focused on AI-driven public services. • Promote Ethical AI:
• Develop guidelines for ethical AI
...moving beyond possibilities www.asti.dost.gov.ph
practices
Philippine Research, Education,
and Government Information
Network (PREGINET)
Davao Ground Receiving Station
PHILSENSORS
Space Technologies
Space Science and Technology Infrastructure

SARwAIS Project
 Systems Engineering
Web and Mobile Applications
KOOHA
NETMESH IoT + Citizen Science

Independent speed test on demand A mobile

application where
NetMesh is a locally developed fixed and mobile
users contribute
benchmarking system used to measure internet
by uploading
performance and quality.
images that can
provide
environmental
parameters and
other
measurements
which could be
used for data
analytics &
research.

Download the KOOHA app on Google Play Store, and capture photos that
go beyond moments!
Embedded Systems Engineering
Advanced Systems Development
arQ 2.0TM
• arQTM is one of ASTI’s field-proven and successfully
commercialized product, which was mass produced and
deployed nationwide

• The 2nd version is being developed to address the following:

o Reliability issues and technical challenges

encountered during the operations of Philsensors
which are equipped with arQTM
o Insufficiency of failure detection, isolation and recovery
features
o Outdated technologies (e.g. 2G/3G comms) and
looming obsolescence of major electronic
components currently being used in the production
assembly
Embedded Systems Engineering
Advanced Systems Development
arQ 2.0TM on Metbuoy

• Various subsystems of arQTM are currently being pilot tested

/demonstrated aboard the meteorological buoy that is
autonomously acquiring 24/7 of weather data and water
quality parameters in real-time at Taal lake near the
vicinity of San Nicolas, Batangas. Deployed last May 2023.
• AI enabled analysis and forecasting on possible occurrence
of fish kill is underway
Embedded Systems Engineering
Advanced Systems Development
Design and Development of a Cost-Effective and Modular
Autonomous System to Trace and Retrieve In-water Information
via an Intelligently Driven Electric Rover (STRIIIDER)
• a real-time telemetering system for water quality
monitoring is essential.
• a safe, cost-effective, and reliable system for more
comprehensive water quality monitoring using a small
Unmanned Surface Vehicle (USV) prototype with an
intuitive ground control software for mission planning,
mapping, and visualization of payload sensor data and
vehicle telemetry data.
• capable of monitoring water quality parameters at
varying depths up to a maximum of ten meters at the
specified multi-location points autonomously.
Ongoing Project Proposals

Monitoring Assets and Quality-Assured Operation

using Intelligent Application (MAQINA) INDUSTRY 4.0
Clinical Decision Support System for Healthcare with AI-Driven

HEALTH Response to Improve Tuberculosis in the Young (CHARITY)

Child-Focus Mapping of Interventions for Neurodevelopmental
Disorders in the Philippines (CHILDMIND)"

Grid Resilience and Intelligent Data Platform for

Automation, Transformation, Control, and
ENERGY
Harmonization - Phase 1 (GRIDPATCH) SECTOR
Content
1. AI in the Philippines & the ASEAN
• Ethical AI Development and Governance
• OneDOST4U! Providing Solutions, Opening Opportunities for the AI Sector through STI
2. DOST-ASTI and its AI Initiatives
• AI projects
• Demos

3. Startup Company Demo