In contexts with limited computational and data resources, high-resource language models often prove inadequate, particularly when addressing the specific needs of Malay languages. This paper introduces a Personal Intelligence System designed to efficiently integrate both on-device and server-based models. The system incorporates SLiM-34M for on-device processing, optimized for low memory and power usage, and MANYAK-1.3B for server-based tasks, allowing for scalable, high-performance language processing. The models achieve significant results across various tasks, such as machine translation, question-answering, and translate IndoMMLU. Particularly noteworthy is SLiM-34M's ability to achieve a high improvement in accuracy compared to other LLMs while using 2 times fewer pre-training tokens. This work challenges the prevailing assumption that large-scale computational resources are necessary to build effective language models, contributing to the development of resource-efficient models for the Malay language with the unique orchestration between SLiM-34M and MANYAK-1.3B.

To meet the requirements of real-world applications, it is essential to control generations of large language models (LLMs). Prior research has tried to introduce reinforcement learning (RL) into controllable text generation while most existing methods suffer from overfitting issues (finetuning-based methods) or semantic collapse (post-processing methods). However, current RL methods are generally guided by coarse-grained (sentence/paragraph-level) feedback, which may lead to suboptimal performance owing to semantic twists or progressions within sentences. To tackle that, we propose a novel reinforcement learning algorithm named TOLE which formulates TOken-LEvel rewards for controllable text generation, and employs a "first-quantize-then-noise" paradigm to enhance the robustness of the RL algorithm.Furthermore, TOLE can be flexibly extended to multiple constraints with little computational expense. Experimental results show that our algorithm can achieve superior performance on both single-attribute and multi-attribute control tasks. We have released our codes at https://github.com/WindyLee0822/CTG

Prompt learning with immensely large Casual Language Models (CLMs) has been shown promising for attribute-controllable text generation (CTG). However, vanilla prompt tuning tends to imitate training corpus characteristics beyond the control attributes, resulting in a poor generalization ability. Moreover, it is less able to capture the relationship between different attributes, further limiting the control performance. In this paper, we propose a new CTG approach, namely DisCup, which incorporates the attribute knowledge of discriminator to optimize the control-prompts, steering a frozen CLM to produce attribute-specific texts. Specifically, the frozen CLM model, capable of producing multitudinous texts, is first used to generate the next-token candidates based on the context, so as to ensure the diversity of tokens to be predicted. Then, we leverage an attribute-discriminator to select desired/undesired tokens from those candidates, providing the inter-attribute knowledge. Finally, we bridge the above two traits by an unlikelihood objective for prompt-tuning. Extensive experimental results show that DisCup can achieve a new state-of-the-art control performance while maintaining an efficient and high-quality text generation, only relying on around 10 virtual tokens.

BANGKOK - An order for mobile phone users in Thailand's restive south to submit a photo of themselves for facial recognition purposes is causing uproar from opponents who see it as further curtailing the rights of the Muslim-majority population. But an army spokesman on Wednesday defended the move, saying the facial identification scheme is needed to root out insurgents deploying mobile phone-detonated home-made bombs. Thailand's three southernmost states -- Yala, Pattani and Narathiwat -- have since 2004 been rife with conflict between Malay-Muslim rebels and the Buddhist-majority Thai state, which annexed the region around a century ago. The tit-for-tat violence has claimed around 7,000 lives, mostly civilians of both faiths, and security forces have detained individuals suspected of being separatist rebels without warrants in the past. Now telecoms companies are requiring all users of the region's 1.5 million mobile numbers to submit a photo of themselves for facial recognition purposes following orders from the army -- a move that is drawing anger from rights groups as the deadline to register photos nears.

Terrorism has led to many problems in Thai societies, not only property damage but also civilian casualties. Predicting terrorism activities in advance can help prepare and manage risk from sabotage by these activities. This paper proposes a framework focusing on event classification in terrorism domain using fuzzy inference systems (FISs). Each FIS is a decision-making model combining fuzzy logic and approximate reasoning. It is generated in five main parts: the input interface, the fuzzification interface, knowledge base unit, decision making unit and output defuzzification interface. Adaptive neuro-fuzzy inference system (ANFIS) is a FIS model adapted by combining the fuzzy logic and neural network. The ANFIS utilizes automatic identification of fuzzy logic rules and adjustment of membership function (MF). Moreover, neural network can directly learn from data set to construct fuzzy logic rules and MF implemented in various applications. FIS settings are evaluated based on two comparisons. The first evaluation is the comparison between unstructured and structured events using the same FIS setting. The second comparison is the model settings between FIS and ANFIS for classifying structured events. The data set consists of news articles related to terrorism events in three southern provinces of Thailand. The experimental results show that the classification performance of the FIS resulting from structured events achieves satisfactory accuracy and is better than the unstructured events. In addition, the classification of structured events using ANFIS gives higher performance than the events using only FIS in the prediction of terrorism events.