DEEP LEARNING
1. INTRODUCTION Dr. Ahmad Nasikun, S.T., M.Sc.
Dr. Ahmad Nasikun, S.T., M.Sc.
Pendidikan
S-1 2007-2012: Teknik Elektro, UGM
S-2 2013-2015: Computer Science and Electrical Engineering
Seoul National University (SNU), Korea Selatan
S-3 2017-2022: Geometry Processing, Compute Graphics
TU Delft, Belanda
Minat Riset
Geometry processing
Deep Learning on Point Clouds
Animation, Deformation, and Simulation
Virtual, Augmented, and Mixed Reality
Email: [email protected]
Virtual Reality for Korean Language Learning
• Interaksi dalam pembelajaran huruf Hangeul
• Pembelajaran fonologi Bahasa Korea
• Analisis fonologi Bahasa Korea dengan machine
learning
WHAT ARE COVERED
Pengenalan tentang Kecerdasan Buatan
Jenis machine learning
Beberapa contoh di industry
Fundamental dari Deep Learning
WHICH FACE IS REAL?
(a) (b) (c)
Source: https://www.theverge.com/tldr/2019/2/15/18226005/ai-g
enerated-fake-people-portraits-thispersondoesnotexist-stylegan
1997 | Deep Blue beats the world champion
TEXT-TO-IMAGE
Membuat gambar dari suatu teks yang dimasukkan oleh pengguna
Input: teks deskripsi (in English)
output: gambar
TEXT-TO-IMAGE
Bisa dengan berbagai gaya gambar
“Indonesian football player playing for Juventus”
TEXT-TO-IMAGE
TEXT-TO-IMAGE
Membuat gambar realistis dari suatu teks yang dimasukkan oleh pengguna
OpenAI Stability DeepAI
Stable
Dall.E Diffusion
https://deepai.org/machine-
learning-model/text2img
https://labs.openai.com/ https://beta.dreamstudio.ai/generat
e
[ADVANCED] CHAT BOX
INTRO ARTIFICIAL INTELLIGENCE
Visual perception
Speech recognition
Decision making
Language translation
….
Creativity
APA PERBEDAANNYA?
WHY NOW?
Big Data Hardware Software
HOW AI LEARNS?
HOW AI LEARNS? (KECOA 0, KURMA 1)
HOW AI LEARNS? (KECOA 0, KURMA 1)
HOW AI LEARNS? (KECOA 0, KURMA 1)
HOW AI LEARNS? (KECOA 0, KURMA 1)
HOW AI LEARNS? (KECOA 0, KURMA 1)
HOW AI LEARNS? (KECOA 0, KURMA 1)
SUPERVISED LEARNING
The model is trained on
labeled data
with the goal of learning to
predict an output variable
based on input variable
EXAMPLE: TESLA SELF-DRIVING CAR
https://www.youtube.com/watch?v=tlThdr3O5Qo
EXAMPLE: AUTONOMOUS DRIVING
UNSUPERVISED LEARNING
Learns patterns from
untagged data
exhibit self-organization
that captures patterns
as probability densities
EXAMPLE ON FACEBOOK
Facebook uses clustering to group users based on their interests, behaviors, and
News feed demographics. If a user has liked a lot of posts related to cats, Facebook will
group them with other users who have similar interests in cats and show them
more cat-related content.
Facebook uses clustering to group users based on their behavior patterns and
Ad targeting demographics, such as age, gender, location, and interests. For example, if an
advertiser wants to target young women who are interested in fitness, Facebook
will group users who have similar interests and show the ads to them.
Friend Facebook uses clustering to suggest friends to users based on their network
connections and behavior patterns. If a user has many friends in common with
suggestions
another user, Facebook's clustering algorithm will suggest that they become
friends.
DETECTING CREDIT CARD FRAUD
CLUSTERING POINT CLOUD DATA
Wiersma, R., Nasikun, A., Eisemann, E., & Hildebrandt, K. (2022). DeltaConv: anisotropic operators for geometric deep learning on point clouds.
ACM Transactions on Graphics (TOG), 41(4), 1-10.
GENERATIVE MODEL
learns the underlying
structure of a dataset
can then generate new
data points similar to
the original data
STYLE-GAN
Karras, T., Laine, S., & Aila, T. (2019). A style-based generator architecture for generative adversarial networks.
In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 4401-4410).
STYLE-GAN
Karras, T., Aittala, M., Laine, S., Härkönen, E., Hellsten, J., Lehtinen, J., & Aila, T. (2021). Alias-free generative adversarial networks.
Advances in Neural Information Processing Systems, 34, 852-863.
Details of
the Network
Forward Pass/Propagation
𝒙 𝒎
𝑚1
𝑥1 𝑚2 𝑻
𝑥2
𝒎=𝑾 𝒙+𝒃
𝑚3
𝑥3
𝑚3
Activation function
𝒎
𝑚1
𝑻
𝑚2 𝒏=𝑾 𝒎
𝑚3 ෝ = 𝒇(𝒏)
𝒚
𝑚3
Loss Function
𝒎
𝑚1
𝑚2 ෝ = 𝒇(𝒏)
𝒚
ෝ, 𝒚)
𝒍𝒐𝒔𝒔 = 𝑳(𝒚
𝑚3
𝑚3
Back propagation
Updating weight
𝑚1
𝑚2 ෝ, 𝒚)
𝒍𝒐𝒔𝒔 = 𝑳(𝒚
𝑚3
𝑚3
TERIMA KASIH Ahmad Nasikun
[email protected]
