Rizal Fathony

Applied Researcher | AI Foundations @ Capital One

[The Cultural Tutor @culturaltutor]

This is Borobudur in Indonesia, one of the world's most important and mysterious buildings.

Why? Because it's the largest Buddhist temple ever built — and it was also abandoned for nearly one thousand years...

[Rizal Fathony @rizalzaf]

We also published an implementation of GraphBEAN and released it as an open-source project. The model is general purpose and is applicable to any bipartite graphs. Please find our implementation on top of @PyTorch and @PyG_Team on our GitHub page. https://t.co/PbEYCTP1Es

[Rizal Fathony @rizalzaf]

As fraudsters continue to evolve, it becomes more challenging to detect new fraudulent behaviors automatically. Find out how we designed a machine learning model that has the ability to discover new fraud patterns without the need for label supervision.
https://t.co/G0sBPCE4NJ

[Rizal Fathony @rizalzaf]

Our novel model was presented at the International Joint Conference on Neural Networks (IJCNN) 2023 conference in Gold Coast, Australia. Please check our paper, "Interaction-Focused Anomaly Detection on Bipartite Node-and-Edge-Attributed Graphs," for the model details.

[Rizal Fathony @rizalzaf]

@ainunnajib Kondisi dimana Indonesia timur masih negatif tp barat sdh positif terjadi paling kalau new moon nya baru di observe <3 jam sebelum magrib. Ini bakal jarang terjadi harusnya.

[Rizal Fathony @rizalzaf]

@ainunnajib Indonesia timur ke barat ~5k km, dibagi ~40k circumference bumi, berarti perbedaan antara timur dan barat Indonesia cuman sekitar 5/40 * 12' = 1.5'
per jam bulan bergerak ~ 12' / 24 = ~0.5'.

[Rizal Fathony @rizalzaf]

@ainunnajib hal yang sama juge berlaku untuk yg lihat hilal mas. kl di sumatra sudah >= 3' (bisa dilihat), tapi Indonesia tengah dan timur masih < 3', gimana keputusannya.

[Rizal Fathony @rizalzaf]

@ainunnajib unless you're warga Muhammadiyah 😃,
new moon = new month

[Rizal Fathony @rizalzaf]

@ainunnajib jadi tidak bisa dijadikan bukti menurut NU.
ya. masing2 punya pendapat sendiri2. saling menghormati saja.

[Rizal Fathony @rizalzaf]

@ainunnajib got it. jadi ini ujung perbedaan dari NU dan muhammadiyah. kl muhamadiyah awal bulan dihitung dari apakah new moon sudah ada saat magrib. kl NU awal bulan dihitung dari apakah bulan sudah di atas curvature bumi (bisa dilihat) pas magrib. gerhana cuma menunjukkan new moon sdh ada.

[Rizal Fathony @rizalzaf]

@ainunnajib Khusus untuk pas gerhana matahari, kita juga bisa melihatnya dengan mata mas, at least shadow bulan baru pas gerhana. Apa ini tidak bisa dipakai juga sebagai bukti?

[Rizal Fathony @rizalzaf]

@ainunnajib pertanyaan nya, apakah ini tidak menjadi cukup bukti 'melihat' bulan baru? kan bisa dilihat secara mata kepala juga? atau apakah melihat bulan baru itu harus pas magrib, dan tidak cukup cuma shadow nya?

[Rizal Fathony @rizalzaf]

@ainunnajib iya mas. kalau pas magrib, karena curvature bumi, hilal membutuhkan lebih dr berapa derajat dl, bari bisa dilihat. kl pas gerhana matahari total, bulan baru bisa kelihatan regardless derajat nya berapa. Bukan bulan nya sih yg kelihatan. Tp shadow matahari yang tertutup bulan

[Rizal Fathony @rizalzaf]

@ainunnajib mas @ainunnajib. Penasaran kalau kasusnya bertepatan dengan gerhana matahari hari ini. Dengan gerhana matahari tadi, bukankah itu juga bisa dipakai sebagai bukti melihat bulan baru? Kan kita bisa melihat dengan teleskop kalau fase bulan baru sudah muncul, walau tdk scr langsung.

[Rizal Fathony @rizalzaf]

@RICEric22 @CIS_Penn Congratulations Eric!

[Vincent Sitzmann @vincesitzmann]

Implicit neural representations have recently gotten a lot of attention. I have compiled a reading list that I give students to get started in this area, inspired by the awesome-computer-vision list with extra commentary & notes. Check it out!

https://t.co/MeMT1jOwB7

[Rizal Fathony @rizalzaf]

8/ Also check our Julia implementation which provides a speed benefit compared with the Python implementation.

https://t.co/OR4y0u7Bjo

[Rizal Fathony @rizalzaf]

1/ We just released a Python implementation of the `AP-Perf` alongside the paper's camera-ready version.

We can now easily optimize neural nets for performance metrics like F1 score and beyond in Python

https://t.co/iXwHPBzocm
https://t.co/rh23EmENuO
Joint work with @zicokolter

[Rizal Fathony @rizalzaf]

7/ The framework also supports writing novel new metrics specifically targeted for new applications. Just augment the `define` and `constraint` methods to apply the framework to new problems.