• LATEST
    • A PHP Error was encountered

      Severity: Notice

      Message: Undefined variable: latest

      Filename: views/_header.php

      Line Number: 48

      A PHP Error was encountered

      Severity: Warning

      Message: Invalid argument supplied for foreach()

      Filename: views/_header.php

      Line Number: 48


Kerangka-Kerja Baru Sistem Intelijen Artifisial

Satu proyek riset kecerdasan buatan (artificial intelligence/AI) tim ahli pada North State University, yang didanai oleh Angkatan Darat Amerika Serikat, menemukan bahwa kerangka kerja baru AI Learn to Grow untuk mempelajari tugas-tugas baru, dapat membuat AI lebih baik melaksanakan tugas-tugas sebelumnya. Fenomena ini disebut ‘backward transfer’ (U.S. Army Research Laboratory, 20/5/2019; CCDC Army Research Laboratory Public Affairs, 20/5/2019).

“In some cases, the Learn to Grow framework actually got better at performing the old tasks. This is called backward transfer, and occurs when you find that learning a new task makes you better at an old task. We see this in people all the time; not so much with AI,” ungkap Caiming Xiong, Ph.D, direktur riset Salesforce Research dan co-author studi ilmiah itu (U.S. Army Research Laboratory, 20/5/2019).

Hasil riset itu membawa teknologi AI selangkah lebih dekat membantu misi-misi militer secara lebih efektif. “This Army investment extends the current state of the art machine learning techniques that will guide our Army Research Laboratory researchers as they develop robotic applications, such as intelligent maneuver and learning to recognize novel objects. This research brings AI a step closer to providing our warfighters with effective unmanned systems that can be deployed in the field,” ungkap Dr. Mary Anne Fields, manajer program  Intelligent Systems pada Army Research Office, satu elemen Army Research Laboratory pada  U.S. Army Combat Capabilities Development Command di Amerika Serikat (U.S. Army Research Laboratory, 20/5/2019).

Lebih rinci Dr. Mary Anne Fields menguraikan kebutuhan sistem intelijen dan pelaksanaan misi berbasis AI. “The Army needs to be prepared to fight anywhere in the world so its intelligent systems also need to be prepared. We expect the Army's intelligent systems to continually acquire new skills as they conduct missions on battlefields around the world without forgetting skills that have already been trained. For instance, while conducting an urban operation, a wheeled robot may learn new navigation parameters for dense urban cities, but it still needs to operate efficiently in a previously encountered environment like a forest,” ungkap Dr. Mary Anne Fields, manajer program  Intelligent Systems pada Army Research Office, satu elemen Army Research Laboratory pada  U.S. Army Combat Capabilities Development Command (U.S. Army Research Laboratory, 20/5/2019).

Tim peneliti itu mengusulkan kerangka kerja baru Learn to Grow untuk pembelajaran berkelanjutan, yang memisahkan pembelajaran struktur jaringan dan pembelajaran parameter model. Dalam uji-coba riset, kerangka kerja baru itu mengungguli pendekatan sebelumnya untuk pembelajaran berkelanjutan (Science Daily, 20/5/2019).

“Deep neural network AI systems are designed for learning narrow tasks. As a result, one of several things can happen when learning new tasks, systems can forget old tasks when learning new ones, which is called catastrophic forgetting. Systems can forget some of the things they knew about old tasks, while not learning to do new ones as well. Or systems can fix old tasks in place while adding new tasks -- which limits improvement and quickly leads to an AI system that is too large to operate efficiently. Continual learning, also called lifelong-learning or learning-to-learn, is trying to address the issue,” ungkap Xilai Li,  co-lead author karya ilmiah itu dan kandidat doktor pada North Carolina State University (NC State) (U.S. Army Research Laboratory, 20/5/2019).

Kerangka Learn to Grow --jaringan-kerja saraf cerdas (deep neural networks)—ibarat pipa terisi banyak lapisan. Data mentah masuk ke bagian atas pipa, dan tugas output keluar dari bagian bawah. Tiap “lapisan” dalam pipa adalah suatu perhitungan yang memanipulasi data untuk membantu jaringan menyelesaikan tugasnya, seperti mengidentifikasi obyek dalam gambar digital. Ada beberapa cara mengatur lapisan dalam pipa, yang sesuai dengan “arsitektur” jaringan berbeda (Science Daily, 20/5/2019).

Learn to Grow  mempelajari tugas baru berawal dari optimasi arsitektur saraf eksplisit melalui pencarian. Yakni ketika jaringan kerja itu datang ke setiap lapisan dalam sistemnya, kerangka itu memilih satu dari empat hal : (1) lewati lapisan; (2) gunakan layer dengan cara sama seperti tugas sebelumnya; (3) pasang adaptor ringan ke lapisan, yang sedikit memodifikasinya; atau (4) membuat layer sama sekali baru. Optimalisasi arsitektur ini memaparkan topologi terbaik, atau serangkaian lapisan menyelesaikan tugas baru. Setelah ini selesai, jaringan menggunakan topologi baru latihan menyelesaikan tugas --sama seperti sistem AI lainnya.

“We've run experiments using several datasets, and what we've found is that the more similar a new task is to previous tasks, the more overlap there is in terms of the existing layers that are kept to perform the new task. What is more interesting is that, with the optimized -- or “learned” topology -- a network trained to perform new tasks forgets very little of what it needed to perform the older tasks, even if the older tasks were not similar,” Xilai Li, mahasiswa doktoral pada NC State (U.S. Army Research Laboratory, 20/5/2019).

Topik studi ilmiah itu “Learn to Grow: A Continual Structure Learning Framework for Overcoming Catastrophic Forgetting” dipresentasikan pada The 36th International Conference on Machine Learning pada 9-15 Juni 2019 di Long Beach, California, Amerika Serikat. Tim riset dan penulis karya ilmiah itu melibatkan satu tim ahli antara lain co-lead author Tianfu Wu, Ph.D., asisten profesor bidang rekayasa komputer dan listrik pada NC State; Xilai Li, mahasiswa doktoral pada NC State, dan Yingbo Zhou dari Salesforce Research; co-author Richard Socher dan Caiming Xiong dari Salesforce Research.

Proyek riset itu didukung oleh Army Research Laboratory, Army Research Office dan National Science Foundation di Amerika Serikat. Sebagian karya ilmiah dilaksanakan oleh Xilai Li ketika masih menjadi bekerja pada Salesforce AI Research di Amerika Serikat. Saat ini CCDC Army Research Laboratory (ARL) adalah elemen pada The U.S. Army Combat Capabilities Development Command. Fokus riset ARL ialah inovasi sains dan teknologi guna menjamin dominasi kekuatan stategis darat militer AS. 

Oleh: Servas Pandur