Ekapol Chuangsuwanich

Wannaphong Phatthiyaphaibun, Chompakorn Chaksangchaichot, Peerat Limkonchotiwat et al.

Recently, Automatic Speech Recognition (ASR), a system that converts audio into text, has caught a lot of attention in the machine learning community. Thus, a lot of publicly available models were released in HuggingFace. However, most of these ASR models are available in English; only a minority of the models are available in Thai. Additionally, most of the Thai ASR models are closed-sourced, and the performance of existing open-sourced models lacks robustness. To address this problem, we train a new ASR model on a pre-trained XLSR-Wav2Vec model with the Thai CommonVoice corpus V8 and train a trigram language model to boost the performance of our ASR model. We hope that our models will be beneficial to individuals and the ASR community in Thailand.

Pitchaporn Rewatbowornwong, Nattanat Chatthee, Ekapol Chuangsuwanich et al.

CLIP has enabled new and exciting joint vision-language applications, one of which is open-vocabulary segmentation, which can locate any segment given an arbitrary text query. In our research, we ask whether it is possible to discover semantic segments without any user guidance in the form of text queries or predefined classes, and label them using natural language automatically? We propose a novel problem zero-guidance segmentation and the first baseline that leverages two pre-trained generalist models, DINO and CLIP, to solve this problem without any fine-tuning or segmentation dataset. The general idea is to first segment an image into small over-segments, encode them into CLIP's visual-language space, translate them into text labels, and merge semantically similar segments together. The key challenge, however, is how to encode a visual segment into a segment-specific embedding that balances global and local context information, both useful for recognition. Our main contribution is a novel attention-masking technique that balances the two contexts by analyzing the attention layers inside CLIP. We also introduce several metrics for the evaluation of this new task. With CLIP's innate knowledge, our method can precisely locate the Mona Lisa painting among a museum crowd. Project page: https://zero-guide-seg.github.io/.

Panuthep Tasawong, Wuttikorn Ponwitayarat, Peerat Limkonchotiwat et al.

Dense retrieval is a basic building block of information retrieval applications. One of the main challenges of dense retrieval in real-world settings is the handling of queries containing misspelled words. A popular approach for handling misspelled queries is minimizing the representations discrepancy between misspelled queries and their pristine ones. Unlike the existing approaches, which only focus on the alignment between misspelled and pristine queries, our method also improves the contrast between each misspelled query and its surrounding queries. To assess the effectiveness of our proposed method, we compare it against the existing competitors using two benchmark datasets and two base encoders. Our method outperforms the competitors in all cases with misspelled queries. Our code and models are available at https://github. com/panuthept/DST-DenseRetrieval.

Peerat Limkonchotiwat, Wuttikorn Ponwitayarat, Lalita Lowphansirikul et al.

Self-supervised sentence representation learning is the task of constructing an embedding space for sentences without relying on human annotation efforts. One straightforward approach is to finetune a pretrained language model (PLM) with a representation learning method such as contrastive learning. While this approach achieves impressive performance on larger PLMs, the performance rapidly degrades as the number of parameters decreases. In this paper, we propose a framework called Self-supervised Cross-View Training (SCT) to narrow the performance gap between large and small PLMs. To evaluate the effectiveness of SCT, we compare it to 5 baseline and state-of-the-art competitors on seven Semantic Textual Similarity (STS) benchmarks using 5 PLMs with the number of parameters ranging from 4M to 340M. The experimental results show that STC outperforms the competitors for PLMs with less than 100M parameters in 18 of 21 cases.

Konpat Preechakul, Sira Sriswasdi, Boonserm Kijsirikul et al.

In medical imaging, Class-Activation Map (CAM) serves as the main explainability tool by pointing to the region of interest. Since the localization accuracy from CAM is constrained by the resolution of the model's feature map, one may expect that segmentation models, which generally have large feature maps, would produce more accurate CAMs. However, we have found that this is not the case due to task mismatch. While segmentation models are developed for datasets with pixel-level annotation, only image-level annotation is available in most medical imaging datasets. Our experiments suggest that Global Average Pooling (GAP) and Group Normalization are the main culprits that worsen the localization accuracy of CAM. To address this issue, we propose Pyramid Localization Network (PYLON), a model for high-accuracy weakly-supervised localization that achieved 0.62 average point localization accuracy on NIH's Chest X-Ray 14 dataset, compared to 0.45 for a traditional CAM model. Source code and extended results are available at https://github.com/cmb-chula/pylon.

Patomporn Payoungkhamdee, Pume Tuchinda, Jinheon Baek et al.

Multi-step reasoning is essential for large language models (LLMs), yet multilingual performance remains challenging. While Chain-of-Thought (CoT) prompting improves reasoning, it struggles with non-English languages due to the entanglement of reasoning and execution. Program-of-Thought (PoT) prompting separates reasoning from execution, offering a promising alternative but shifting the challenge to generating programs from non-English questions. We propose a framework to evaluate PoT by separating multilingual reasoning from code execution to examine (i) the impact of fine-tuning on question-reasoning alignment and (ii) how reasoning quality affects answer correctness. Our findings demonstrate that PoT fine-tuning substantially enhances multilingual reasoning, outperforming CoT fine-tuned models. We further demonstrate a strong correlation between reasoning quality (measured through code quality) and answer accuracy, highlighting its potential as a test-time performance improvement heuristic.

Napat Laosaengpha, Thanit Tativannarat, Attapol Rutherford et al.

Understanding the textual components of resumes and job postings is critical for improving job-matching accuracy and optimizing job search systems in online recruitment platforms. However, existing works primarily focus on analyzing individual components within this information, requiring multiple specialized tools to analyze each aspect. Such disjointed methods could potentially hinder overall generalizability in recruitment-related text processing. Therefore, we propose a unified sentence encoder that utilized multi-task dual-encoder framework for jointly learning multiple component into the unified sentence encoder. The results show that our method outperforms other state-of-the-art models, despite its smaller model size. Moreover, we propose a novel metric, Language Bias Kullback-Leibler Divergence (LBKL), to evaluate language bias in the encoder, demonstrating significant bias reduction and superior cross-lingual performance.

Wannaphong Phatthiyaphaibun, Can Udomcharoenchaikit, Pakpoom Singkorapoom et al.

Pre-training data shapes a language model's quality, but raw web text is noisy and demands careful cleaning. Existing large-scale corpora rely on English-centric or language-agnostic pipelines whose heuristics do not capture Thai script or cultural nuances, leaving risky material such as gambling content untreated. Prior Thai-specific efforts customize pipelines or build new ones, yet seldom release their data or document design choices, hindering reproducibility and raising the question of how to construct a transparent, high-quality Thai corpus. We introduce Mangosteen: a 47 billion-token Thai corpus built through a Thai-adapted Dolma pipeline that includes custom rule-based language ID, revised C4/Gopher quality filters, and Thai-trained content filters, plus curated non-web sources such as Wikipedia, Royal Gazette texts, OCR-extracted books, and CC-licensed YouTube subtitles. Systematic ablations using GPT-2 show the pipeline trims CommonCrawl from 202M to 25M documents while raising SEA-HELM NLG from 3 to 11; an 8B-parameter SEA-LION model continually pre-trained on Mangosteen then surpasses SEA-LION-v3 and Llama-3.1 by about four points on Thai benchmarks. We release the full pipeline code, cleaning manifests, corpus snapshot, and all checkpoints, providing a fully reproducible foundation for future Thai and regional LLM research.

Muhammad Dehan Al Kautsar, Aswin Candra, Muhammad Alif Al Hakim et al.

Although numerous datasets have been developed to support dialogue systems, most existing chit-chat datasets overlook the cultural nuances inherent in natural human conversations. To address this gap, we introduce SEADialogues, a culturally grounded dialogue dataset centered on Southeast Asia, a region with over 700 million people and immense cultural diversity. Our dataset features dialogues in eight languages from six Southeast Asian countries, many of which are low-resource despite having sizable speaker populations. To enhance cultural relevance and personalization, each dialogue includes persona attributes and two culturally grounded topics that reflect everyday life in the respective communities. Furthermore, we release a multi-turn dialogue dataset to advance research on culturally aware and human-centric large language models, including conversational dialogue agents.

Pawitsapak Akarajaradwong, Chompakorn Chaksangchaichot, Pirat Pothavorn et al.

The Retrieval-Augmented Generation (RAG) systems' performance on Thai legal question answering is still limited, especially for questions requiring extensive, complex legal reasoning. To address these limitations, we introduce an approach aligning LLMs toward improved law citation accuracy and better response quality using Group-Relative Policy Optimization (GRPO). Our approach leverages BGE-M3 embeddings as a cost-efficient semantic-similarity reward, significantly reducing computational expenses up to 2.5x compared to large language model judges. Experiments on the NitiBench benchmark demonstrate substantial improvements: GRPO achieves up to 90% citation-F1 gains from the base model and a 31% increase in joint quality metrics over instruction tuning. Crucially, our method shows enhanced robustness on complex legal reasoning tasks compared to instruction tuning, providing an effective and resource-efficient solution for enhancing Thai legal LLMs.

Patawee Prakrankamanant, Shinji Watanabe, Ekapol Chuangsuwanich

This paper addresses the critical need for improved explainability in text-based depression detection. While offering predictive outcomes, current solutions often overlook the understanding of model predictions which can hinder trust in the system. We propose the use of Masked Hard Instance Mining (MHIM) to enhance the explainability in the depression detection task. MHIM strategically masks attention weights within the model, compelling it to distribute attention across a wider range of salient features. We evaluate MHIM on two datasets representing distinct languages: Thai (Thai-Maywe) and English (DAIC-WOZ). Our results demonstrate that MHIM significantly improves performance in terms of both prediction accuracy and explainability metrics.

Yuatyong Chaichana, Thanapat Trachu, Peerat Limkonchotiwat et al. · berkeley

In the era of large-scale training, model merging has evolved into a tool for creating multitasking models efficiently. It enables the knowledge of models to be fused, without the need for heavy computation as required in traditional multitask learning. Existing merging methods often assume that entries at identical positions in weight matrices serve the same function, enabling straightforward entry-wise comparison and merging. However, this assumption overlooks the complexity of finetuned neural networks, where neurons may develop distinct feature compositions, making direct entry-wise merging problematic. We present Decom-Renorm-Merge (DRM), a simple yet effective approach that leverages Singular Value Decomposition to decompose and coordinate weight matrices into an aligned joint space, where entry-wise merging becomes possible. We showcase the effectiveness of DRM across various settings ranging from smaller encoder-based such as ViT and DeBERTa, encoder-decoder-based such as T5, and larger decoder-based such as Llama3.1-8B. Our experimental results show that DRM outperforms several state-of-the-art merging techniques across full finetuning and low-rank adaptation settings. Moreover, our analysis reveals renormalization as the crucial component for creating a robust and even joint space for merging, significantly contributing to the method's performance.

Napat Laosaengpha, Thanit Tativannarat, Chawan Piansaddhayanon et al.

Learning job title representation is a vital process for developing automatic human resource tools. To do so, existing methods primarily rely on learning the title representation through skills extracted from the job description, neglecting the rich and diverse content within. Thus, we propose an alternative framework for learning job titles through their respective job description (JD) and utilize a Job Description Aggregator component to handle the lengthy description and bidirectional contrastive loss to account for the bidirectional relationship between the job title and its description. We evaluated the performance of our method on both in-domain and out-of-domain settings, achieving a superior performance over the skill-based approach.

Thanapat Trachu, Chawan Piansaddhayanon, Ekapol Chuangsuwanich

Diffusion-based speech enhancement has shown promising results, but can suffer from a slower inference time. Initializing the diffusion process with the enhanced audio generated by a regression-based model can be used to reduce the computational steps required. However, these approaches often necessitate a regression model, further increasing the system's complexity. We propose Thunder, a unified regression-diffusion model that utilizes the Brownian bridge process which can allow the model to act in both modes. The regression mode can be accessed by setting the diffusion time step closed to 1. However, the standard score-based diffusion modeling does not perform well in this setup due to gradient instability. To mitigate this problem, we modify the diffusion model to predict the clean speech instead of the score function, achieving competitive performance with a more compact model size and fewer reverse steps.

Danupat Khamnuansin, Tawunrat Chalothorn, Ekapol Chuangsuwanich

Large Language Models (LLMs) often struggle with hallucinations and outdated information. To address this, Information Retrieval (IR) systems can be employed to augment LLMs with up-to-date knowledge. However, existing IR techniques contain deficiencies, posing a performance bottleneck. Given the extensive array of IR systems, combining diverse approaches presents a viable strategy. Nevertheless, prior attempts have yielded restricted efficacy. In this work, we propose an approach that leverages learning-to-rank techniques to combine heterogeneous IR systems. We demonstrate the method on two Retrieval Question Answering (ReQA) tasks. Our empirical findings exhibit a significant performance enhancement, outperforming previous approaches and achieving state-of-the-art results on ReQA SQuAD.

Wuttikorn Ponwitayarat, Peerat Limkonchotiwat, Ekapol Chuangsuwanich et al.

Determining sentence pair similarity is crucial for various NLP tasks. A common technique to address this is typically evaluated on a continuous semantic textual similarity scale from 0 to 5. However, based on a linguistic observation in STS annotation guidelines, we found that the score in the range [4,5] indicates an upper-range sample, while the rest are lower-range samples. This necessitates a new approach to treating the upper-range and lower-range classes separately. In this paper, we introduce a novel embedding space decomposition method called MixSP utilizing a Mixture of Specialized Projectors, designed to distinguish and rank upper-range and lower-range samples accurately. The experimental results demonstrate that MixSP decreased the overlap representation between upper-range and lower-range classes significantly while outperforming competitors on STS and zero-shot benchmarks.

Chawan Piansaddhayanon, Sakun Santisukwongchote, Shanop Shuangshoti et al.

Mitotic count (MC) is an important histological parameter for cancer diagnosis and grading, but the manual process for obtaining MC from whole-slide histopathological images is very time-consuming and prone to error. Therefore, deep learning models have been proposed to facilitate this process. Existing approaches utilize a two-stage pipeline: the detection stage for identifying the locations of potential mitotic cells and the classification stage for refining prediction confidences. However, this pipeline formulation can lead to inconsistencies in the classification stage due to the poor prediction quality of the detection stage and the mismatches in training data distributions between the two stages. In this study, we propose a Refine Cascade Network (ReCasNet), an enhanced deep learning pipeline that mitigates the aforementioned problems with three improvements. First, window relocation was used to reduce the number of poor quality false positives generated during the detection stage. Second, object re-cropping was performed with another deep learning model to adjust poorly centered objects. Third, improved data selection strategies were introduced during the classification stage to reduce the mismatches in training data distributions. ReCasNet was evaluated on two large-scale mitotic figure recognition datasets, canine cutaneous mast cell tumor (CCMCT) and canine mammary carcinoma (CMC), which resulted in up to 4.8% percentage point improvements in the F1 scores for mitotic cell detection and 44.1% reductions in mean absolute percentage error (MAPE) for MC prediction. Techniques that underlie ReCasNet can be generalized to other two-stage object detection networks and should contribute to improving the performances of deep learning models in broad digital pathology applications.

Burin Naowarat, Thananchai Kongthaworn, Korrawe Karunratanakul et al.

Code-Switching (CS) remains a challenge for Automatic Speech Recognition (ASR), especially character-based models. With the combined choice of characters from multiple languages, the outcome from character-based models suffers from phoneme duplication, resulting in language-inconsistent spellings. We propose Contextualized Connectionist Temporal Classification (CCTC) loss to encourage spelling consistencies of a character-based non-autoregressive ASR which allows for faster inference. The CCTC loss conditions the main prediction on the predicted contexts to ensure language consistency in the spellings. In contrast to existing CTC-based approaches, CCTC loss does not require frame-level alignments, since the context ground truth is obtained from the model's estimated path. Compared to the same model trained with regular CTC loss, our method consistently improved the ASR performance on both CS and monolingual corpora.

Chanatip Saetia, Ekapol Chuangsuwanich, Tawunrat Chalothorn et al.

A sentence is typically treated as the minimal syntactic unit used for extracting valuable information from a longer piece of text. However, in written Thai, there are no explicit sentence markers. We proposed a deep learning model for the task of sentence segmentation that includes three main contributions. First, we integrate n-gram embedding as a local representation to capture word groups near sentence boundaries. Second, to focus on the keywords of dependent clauses, we combine the model with a distant representation obtained from self-attention modules. Finally, due to the scarcity of labeled data, for which annotation is difficult and time-consuming, we also investigate and adapt Cross-View Training (CVT) as a semi-supervised learning technique, allowing us to utilize unlabeled data to improve the model representations. In the Thai sentence segmentation experiments, our model reduced the relative error by 7.4% and 10.5% compared with the baseline models on the Orchid and UGWC datasets, respectively. We also applied our model to the task of pronunciation recovery on the IWSLT English dataset. Our model outperformed the prior sequence tagging models, achieving a relative error reduction of 2.5%. Ablation studies revealed that utilizing n-gram presentations was the main contributing factor for Thai, while the semi-supervised training helped the most for English.

Patcharin Cheng, Phairot Autthasan, Boriwat Pijarana et al.

In this paper, the deep learning (DL) approach is applied to a joint training scheme for asynchronous motor imagery-based Brain-Computer Interface (BCI). The proposed DL approach is a cascade of one-dimensional convolutional neural networks and fully-connected neural networks (CNN-FC). The focus is mainly on three types of brain responses: non-imagery EEG (\textit{background EEG}), (\textit{pure imagery}) EEG, and EEG during the transitional period between background EEG and pure imagery (\textit{transitional imagery}). The study of transitional imagery signals should provide greater insight into real-world scenarios. It may be inferred that pure imagery and transitional EEG are high and low power EEG imagery, respectively. Moreover, the results from the CNN-FC are compared to the conventional approach for motor imagery-BCI, namely the common spatial pattern (CSP) for feature extraction and support vector machine (SVM) for classification (CSP-SVM). Under a joint training scheme, pure and transitional imagery are treated as the same class, while background EEG is another class. Ten-fold cross-validation is used to evaluate whether the joint training scheme significantly improves the performance task of classifying pure and transitional imagery signals from background EEG. Using sparse of just a few electrode channels ($C_{z}$, $C_{3}$ and $C_{4}$), mean accuracy reaches 71.52 % and 70.27 % for CNN-FC and CSP-SVM, respectively. On the other hand, mean accuracy without the joint training scheme achieve only 62.68 % and 52.41 % for CNN-FC and CSP-SVM, respectively.

Itthi Chatnuntawech, Kittipong Tantisantisom, Paisan Khanchaitit et al.

Rice has been one of the staple foods that contribute significantly to human food supplies. Numerous rice varieties have been cultivated, imported, and exported worldwide. Different rice varieties could be mixed during rice production and trading. Rice impurities could damage the trust between rice importers and exporters, calling for the need to develop a rice variety inspection system. In this work, we develop a non-destructive rice variety classification system that benefits from the synergy between hyperspectral imaging and deep convolutional neural network (CNN). The proposed method uses a hyperspectral imaging system to simultaneously acquire complementary spatial and spectral information of rice seeds. The rice varieties are then determined from the acquired spatio-spectral data using a deep CNN. As opposed to several existing rice variety classification methods that require hand-engineered features, the proposed method automatically extracts spatio-spectral features from the raw sensor data. As demonstrated using two types of rice datasets, the proposed method achieved up to 11.9% absolute improvement in the mean classification accuracy, compared to the commonly used classification methods based on support vector machines.

Yu Zhang, Ekapol Chuangsuwanich, James Glass et al.

In this paper, we investigate the use of prediction-adaptation-correction recurrent neural networks (PAC-RNNs) for low-resource speech recognition. A PAC-RNN is comprised of a pair of neural networks in which a {\it correction} network uses auxiliary information given by a {\it prediction} network to help estimate the state probability. The information from the correction network is also used by the prediction network in a recurrent loop. Our model outperforms other state-of-the-art neural networks (DNNs, LSTMs) on IARPA-Babel tasks. Moreover, transfer learning from a language that is similar to the target language can help improve performance further.