Oceania
SaulLM-54B & SaulLM-141B: Scaling Up Domain Adaptation for the Legal Domain
The integration of synthetically generated data in the second and third steps enhances the models' capabilities in interpreting and processing legal texts, effectively reaching state-of-the-art performance and outperforming
How hackers forced brewing giant Asahi back to pen and paper
Only four bottles of Asahi Super Dry beer are left on the shelves of Ben Thai, a cosy restaurant in the Tokyo suburb of Sengawacho. Its owner, Sakaolath Sugizaki, expects to get a few more soon, but she says her supplier is keeping the bulk of its stock for bigger customers. That's because Asahi, the maker of Japan's best-selling beer, was forced to halt production at most of its 30 factories in the country at the end of last month after being hit by a cyber-attack. While all of its facilities in Japan - including six breweries - have now partially reopened, its computer systems are still down. That means it has to process orders and shipments manually - using pen, paper and fax machines - resulting in much fewer shipments than before the attack.
Instance-Optimal Private Density Estimation in the Wasserstein Distance
Estimating the density of a distribution from samples is a fundamental problem in statistics. In many practical settings, the Wasserstein distance is an appropriate error metric for density estimation. For example, when estimating population densities in a geographic region, a small Wasserstein distance means that the estimate is able to capture roughly where the population mass is. In this work we study differentially private density estimation in the Wasserstein distance. We design and analyze instance-optimal algorithms for this problem that can adapt to easy instances.
Provable Partially Observable Reinforcement Learning with Privileged Information Yang Cai
Partial observability of the underlying states generally presents significant challenges for reinforcement learning (RL).
Exploiting Descriptive Completeness Prior for Cross Modal Hashing with Incomplete Labels
As shown in Figure 1, negative pairs are completely removed even with 35% unknown labels in Flickr dataset.
Zero-Shot Tokenizer Transfer Benjamin Minixhofer
Our method comes close to the original models' performance in cross-lingual and
Revealing Distribution Discrepancy by Sampling Transfer in Unlabeled Data
The assumption that data are independently and identically distributed (IID) is staple in statistical machine learning. It suggests that a hypothesis selected by an algorithm, after observing several training samples, should perform effectively on test samples from the same unknown distribution.
PEACE: A Dataset of Pharmaceutical Care for Cancer Pain Analgesia Evaluation and Medication Decision
Over half of cancer patients experience long-term pain management challenges.