Badr M. Abdullah, PhD 🇾🇪
1,908
posts
@badr_nlp
Researcher
@LSTSaar| Saarland University🦉 💬 Speech & Language Processing 💫 Machine Learning 🤖 Cognitive Science 🧠
Berlin, Germany
Joined December 2017
Pinned
🚨 Speech models such as wav2vec 2.0 includes a q-module discretizing speech signal ➡️ discrete units 🟣🟢🟡 But how to characterize these units? In our
@ISCAInterspeechpaper, we develop an information-theoretic framework to this end Paper 📜 arxiv.org/pdf/2306.02405… 1/🧵
🤣
I find this paper insightful But how often we know the target domain data when pretraining LLMs?
Models are typically specialized to new domains by finetuning on small, high-quality datasets. We find that repeating the same dataset 10–50× starting from pretraining leads to substantially better downstream performance, in some cases outperforming larger models. 🧵
Fascinating! Mehri is a South Arabian language and it's completely unintelligible to other semetic languages including Arabic Other South Arabian languages are extinct, but the exception of Shehri in Oman and Socotri in Socotra Island, Yemen
Listen to Mehri language Mehri people are one of a few indigenous groups in the Southern Arabian Peninsula who still speak their own native South Semitic language استمع إلى اللغة المهرية، إحدى اللغات السامية الجنوبية الأصيلة في جنوب شبه الحزيرة
00:00
LOL It would have been more accurate if Mamdani required school students to learn Indo-Arabic numerals
BREAKING: Zohran Mamdani is expected to require ALL New York Elementary school students to learn Arabic numerals. As a Jewish American I still support this 100%
