Chemical representation learning has gained increasing interest due to the limited availability of supervised data in fields such as drug and materials design. This interest particularly extends to chemical language representation learning, which involves pre-training Transformers on SMILES sequences - textual descriptors of molecules. Despite its success in molecular property prediction, current practices often lead to overfitting and limited scalability due to early convergence. In this paper, we introduce a novel chemical language representation learning framework, called MolTRES, to address these issues. MolTRES incorporates generator-discriminator training, allowing the model to learn from more challenging examples that require structural understanding. In addition, we enrich molecular representations by transferring knowledge from scientific literature by integrating external materials embedding. Experimental results show that our model outperforms existing state-of-the-art models on popular molecular property prediction tasks.

Large Language Models (LLMs) have displayed remarkable performances across various complex tasks by leveraging Chain-of-Thought (CoT) prompting. Recently, studies have proposed a Knowledge Distillation (KD) approach, reasoning distillation, which transfers such reasoning ability of LLMs through fine-tuning language models of multi-step rationales generated by LLM teachers. However, they have inadequately considered two challenges regarding insufficient distillation sets from the LLM teacher model, in terms of 1) data quality and 2) soft label provision. In this paper, we propose Mentor-KD, which effectively distills the multi-step reasoning capability of LLMs to smaller LMs while addressing the aforementioned challenges. Specifically, we exploit a mentor, intermediate-sized task-specific fine-tuned model, to augment additional CoT annotations and provide soft labels for the student model during reasoning distillation. We conduct extensive experiments and confirm Mentor-KD’s effectiveness across various models and complex reasoning tasks.

Pre-trained chemical language models (CLMs) excel in the field of molecular property prediction, utilizing string-based molecular descriptors such as SMILES for learning universal representations. However, such string-based descriptors implicitly contain limited structural information, which is closely associated with molecular property prediction. In this work, we introduce Moleco, a novel contrastive learning framework to enhance the understanding of molecular structures within CLMs. Based on the similarity of fingerprint vectors among different molecules, we train CLMs to distinguish structurally similar and dissimilar molecules in a contrastive manner. Experimental results demonstrate that Moleco significantly improves the molecular property prediction performance of CLMs, outperforming state-of-the-art models. Moreover, our in-depth analysis with diverse Moleco variants verifies that fingerprint vectors are highly effective features in improving CLMs’ understanding of the structural information of molecules.

Materials science is an interdisciplinary field focused on studying and discovering materials around us. However, due to the vast space of materials, datasets in this field are typically scarce and have limited coverage. This inherent limitation makes current adaptation methods less effective when adapting pre-trained language models (PLMs) to materials science, as these methods rely heavily on the frequency information from limited downstream datasets. In this paper, we propose Semantic Knowledge Transfer (SEED), a novel vocabulary expansion method to adapt the pre-trained language models for materials science. The core strategy of SEED is to transfer the materials knowledge of lightweight embeddings into the PLMs. To this end, we introduce knowledge bridge networks, which learn to transfer the latent knowledge of the materials embeddings into ones compatible with PLMs. By expanding the embedding layer of PLMs with these transformed embeddings, PLMs can comprehensively understand the complex terminology associated with materials science. We conduct extensive experiments across a broad range of materials-related benchmarks. Comprehensive evaluation results convincingly demonstrate that SEED mitigates the mentioned limitations of previous adaptation methods, showcasing the efficacy of transferring embedding knowledge into PLMs.

The Korean writing system, Hangeul, has a unique character representation rigidly following the invention principles recorded in Hunminjeongeum. However, existing pre-trained language models (PLMs) for Korean have overlooked these principles. In this paper, we introduce a novel framework for Korean PLMs called KOMBO, which firstly brings the invention principles of Hangeul to represent character. Our proposed method, KOMBO, exhibits notable experimental proficiency across diverse NLP tasks. In particular, our method outperforms the state-of-the-art Korean PLM by an average of 2.11% in five Korean natural language understanding tasks. Furthermore, extensive experiments demonstrate that our proposed method is suitable for comprehending the linguistic features of the Korean language. Consequently, we shed light on the superiority of using subcharacters over the typical subword-based approach for Korean PLMs. Our code is available at: https://github.com/SungHo3268/KOMBO.

In this paper, we introduce COCONUT to effectively guide the contextualization of structured commonsense knowledge based on largelanguage models. COCONUT employs a contextualized knowledge prompting scheme to gather high-quality contextualization examplesfrom a large language model. These examples are subsequently distilled into small language models to enhance their contextualization capability. Extensive evaluations show that COCONUT considerably improves commonsense reasoning performance across diverse benchmarks, models, and settings, exhibiting its flexibility and universality in generating contextualized commonsense knowledge. Notably,COCONUT consistently outperforms the state-of-the-art technique by an average of 5.8%.

We introduce a novel continued pre-training method, MELT (MatEriaLs-aware continued pre-Training), specifically designed to efficiently adapt the pre-trained language models (PLMs) for materials science. Unlike previous adaptation strategies that solely focus on constructing domain-specific corpus, MELT comprehensively considers both the corpus and the training strategy, given that materials science corpus has distinct characteristics from other domains. To this end, we first construct a comprehensive materials knowledge base from the scientific corpus by building semantic graphs. Leveraging this extracted knowledge, we integrate a curriculum into the adaptation process that begins with familiar and generalized concepts and progressively moves toward more specialized terms. We conduct extensive experiments across diverse benchmarks to verify the effectiveness and generality of MELT. A comprehensive evaluation convincingly supports the strength of MELT, demonstrating superior performance compared to existing continued pre-training methods. In-depth analysis also shows that MELT enables PLMs to effectively represent materials entities compared to the existing adaptation methods, thereby highlighting its broad applicability across a wide spectrum of materials science.

Recent approaches to zero-shot commonsense reasoning have enabled Pre-trained Language Models (PLMs) to learn a broad range of commonsense knowledge without being tailored to specific situations. However, they often suffer from human reporting bias inherent in textual commonsense knowledge, leading to discrepancies in understanding between PLMs and humans. In this work, we aim to bridge this gap by introducing an additional information channel to PLMs. We propose Imagine (Machine Imagination-based Reasoning), a novel zero-shot commonsense reasoning framework designed to complement textual inputs with visual signals derived from machine-generated images. To achieve this, we enhance PLMs with imagination capabilities by incorporating an image generator into the reasoning process. To guide PLMs in effectively leveraging machine imagination, we create a synthetic pre-training dataset that simulates visual question-answering. Our extensive experiments on diverse reasoning benchmarks and analysis show that Imagine outperforms existing methods by a large margin, highlighting the strength of machine imagination in mitigating reporting bias and enhancing generalization capabilities.

Parameter-efficient fine-tuning (PEFT) has enabled the efficient optimization of cumbersome language models in real-world settings. However, as datasets in such environments often contain noisy labels that adversely affect performance, PEFT methods are inevitably exposed to noisy labels. Despite this challenge, the adaptability of PEFT to noisy environments remains underexplored. To bridge this gap, we investigate various PEFT methods under noisy labels. Interestingly, our findings reveal that PEFT has difficulty in memorizing noisy labels due to its inherently limited capacity, resulting in robustness. However, we also find that such limited capacity simultaneously makes PEFT more vulnerable to interference of noisy labels, impeding the learning of clean samples. To address this issue, we propose Clean Routing (CleaR), a novel routing-based PEFT approach that adaptively activates PEFT modules. In CleaR, PEFT modules are preferentially exposed to clean data while bypassing the noisy ones, thereby minimizing the noisy influence. To verify the efficacy of CleaR, we perform extensive experiments on diverse configurations of noisy labels. The results convincingly demonstrate that CleaR leads to substantially improved performance in noisy environments

Transformers have become the de-facto standard for natural language processing. However, dense information flows within transformers pose significant challenges for real-time and resource-constrained devices, as computational complexity grows quadratically with sequence length. To counteract such dense information flows, we propose SparseFlow, a novel efficient method designed to sparsify the dense pathways of token representations across all transformer blocks. To this end, SparseFlow parameterizes the information flows linking token representations to transformer blocks. These parameterized information flows are optimized to be sparse, allowing only the salient information to pass through into the blocks. To validate the efficacy of SparseFlow, we conduct comprehensive experiments across diverse benchmarks (understanding and generation), scales (ranging from millions to billions), architectures (including encoders, decoders, and seq-to-seq models), and modalities (such as language-only and vision-language). The results convincingly demonstrate that sparsifying the dense information flows leads to substantial speedup gains without compromising task accuracy. For instance, SparseFlow reduces computational costs by half on average, without a significant loss in accuracy.

Warning: This paper contains some offensive expressions. Offensive content is an unavoidable issue on social media. Most existing offensive language identification methods rely on the compilation of labeled datasets. However, existing methods rarely consider low-resource languages that have relatively less data available for training (e.g., Korean). To address these issues, we construct a novel KOrean Dataset for Offensive Language Identification (KODOLI). KODOLI comprises more fine-grained offensiveness categories (i.e., not offensive, likely offensive, and offensive) than existing ones. A likely offensive language refers to texts with implicit offensiveness or abusive language without offensive intentions. In addition, we propose two auxiliary tasks to help identify offensive languages: abusive language detection and sentiment analysis. We provide experimental results for baselines on KODOLI and observe that language models suffer from identifying “LIKELY” offensive statements. Quantitative results and qualitative analysis demonstrate that jointly learning offensive language, abusive language and sentiment information improves the performance of offensive language identification.

Despite the versatility of pre-trained language models (PLMs) across domains, their large memory footprints pose significant challenges in federated learning (FL), where the training model has to be distributed between a server and clients. One potential solution to bypass such constraints might be the use of parameter-efficient fine-tuning (PEFT) in the context of FL. However, we have observed that typical PEFT tends to severely suffer from heterogeneity among clients in FL scenarios, resulting in unstable and slow convergence. In this paper, we propose Client-Customized Adaptation (C2A), a novel hypernetwork-based FL framework that generates client-specific adapters by conditioning the client information. With the effectiveness of the hypernetworks in generating customized weights through learning to adopt the different characteristics of inputs, C2A can maximize the utility of shared model parameters while minimizing the divergence caused by client heterogeneity. To verify the efficacy of C2A, we perform extensive evaluations on FL scenarios involving heterogeneity in label and language distributions. Comprehensive evaluation results clearly support the superiority of C2A in terms of both efficiency and effectiveness in FL scenarios.

Towards human-level visual understanding, visual commonsense generation has been introduced to generate commonsense inferences beyond images. However, current research on visual commonsense generation has overlooked an important human cognitive ability: generating descriptive and diverse inferences. In this work, we propose a novel visual commonsense generation framework, called DIVE, which aims to improve the descriptiveness and diversity of generated inferences. DIVE involves two methods, generic inference filtering and contrastive retrieval learning, which address the limitations of existing visual commonsense resources and training objectives. Experimental results verify that DIVE outperforms state-of-the-art models for visual commonsense generation in terms of both descriptiveness and diversity, while showing a superior quality in generating unique and novel inferences. Notably, DIVE achieves human-level descriptiveness and diversity on Visual Commonsense Graphs. Furthermore, human evaluations confirm that DIVE aligns closely with human judgments on descriptiveness and diversity.

Biases in the dataset often enable the model to achieve high performance on in-distribution data, while poorly performing on out-of-distribution data. To mitigate the detrimental effect of the bias on the networks, previous works have proposed debiasing methods that down-weight the biased examples identified by an auxiliary model, which is trained with explicit bias labels. However, finding a type of bias in datasets is a costly process. Therefore, recent studies have attempted to make the auxiliary model biased without the guidance (or annotation) of bias labels, by constraining the model’s training environment or the capability of the model itself. Despite the promising debiasing results of recent works, the multi-class learning objective, which has been naively used to train the auxiliary model, may harm the bias mitigation effect due to its regularization effect and competitive nature across classes. As an alternative, we propose a new debiasing framework that introduces binary classifiers between the auxiliary model and the main model, coined bias experts. Specifically, each bias expert is trained on a binary classification task derived from the multi-class classification task via the One-vs-Rest approach. Experimental results demonstrate that our proposed strategy improves the bias identification ability of the auxiliary model. Consequently, our debiased model consistently outperforms the state-of-the-art on various challenge datasets.

Prompt tuning has emerged as a successful parameter-efficient alternative to the full fine-tuning of language models. However, prior works on prompt tuning often utilize long soft prompts of up to 100 tokens to improve performance, overlooking the inefficiency associated with extended inputs. In this paper, we propose a novel prompt tuning method SMoP (Sparse Mixture-of-Prompts) that utilizes short soft prompts for efficient training and inference while maintaining performance gains typically induced from longer soft prompts. To achieve this, SMoP employs a gating mechanism to train multiple short soft prompts specialized in handling different subsets of the data, providing an alternative to relying on a single long soft prompt to cover the entire data. Experimental results demonstrate that SMoP outperforms baseline methods while reducing training and inference costs. We release our code at https://github.com/jyjohnchoi/SMoP.

Computational inefficiency in transformers has been a long-standing challenge, hindering the deployment in resource-constrained or real-time applications. One promising approach to mitigate this limitation is to progressively remove less significant tokens, given that the sequence length strongly contributes to the inefficiency. However, this approach entails a potential risk of losing crucial information due to the irrevocable nature of token removal. In this paper, we introduce Leap-of-Thought (LoT), a novel token reduction approach that dynamically routes tokens within layers. Unlike previous work that irrevocably discards tokens, LoT enables tokens to ‘leap’ across layers. This ensures that all tokens remain accessible in subsequent layers while reducing the number of tokens processed within layers. We achieve this by pairing the transformer with dynamic token routers, which learn to selectively process tokens essential for the task. Evaluation results clearly show that LoT achieves a substantial improvement in computational efficiency. Specifically, LoT attains up to 25x faster inference time without a significant loss in accuracy

We introduce Basic, Tiniest Subword (BTS) units for the Korean language, which are inspired by the invention principle of Hangeul, the Korean writing system. Instead of relying on 51 Korean consonant and vowel letters, we form the letters from BTS units by adding strokes or combining them. To examine the impact of BTS units on Korean language processing, we develop a novel BTS-based word embedding framework that is readily applicable to various models. Our experiments reveal that BTS units significantly improve the performance of Korean word embedding on all intrinsic and extrinsic tasks in our evaluation. In particular, BTS-based word embedding outperforms the state-of-theart Korean word embedding by 11.8% in word analogy. We further investigate the unique advantages provided by BTS units through indepth analysis.

Pre-trained language models have achieved remarkable successes in natural language processing tasks, coming at the cost of increasing model size. To address this issue, knowledge distillation (KD) has been widely applied to compress language models. However, typical KD approaches for language models have overlooked the difficulty of training examples, suffering from incorrect teacher prediction transfer and sub-efficient training. In this paper, we propose a novel KD framework, Tutor-KD, which improves the distillation effectiveness by controlling the difficulty of training examples during pre-training. We introduce a tutor network that generates samples that are easy for the teacher but difficult for the student, with training on a carefully designed policy gradient method. Experimental results show that Tutor-KD significantly and consistently outperforms the state-of-the-art KD methods with variously sized student models on the GLUE benchmark, demonstrating that the tutor can effectively generate training examples for the student.

Self-supervised pre-training has achieved remarkable success in extensive natural language processing tasks. Masked language modeling (MLM) has been widely used for pre-training effective bidirectional representations but comes at a substantial training cost. In this paper, we propose a novel concept-based curriculum masking (CCM) method to efficiently pre-train a language model. CCM has two key differences from existing curriculum learning approaches to effectively reflect the nature of MLM. First, we introduce a novel curriculum that evaluates the MLM difficulty of each token based on a carefully-designed linguistic difficulty criterion. Second, we construct a curriculum that masks easy words and phrases first and gradually masks related ones to the previously masked ones based on a knowledge graph. Experimental results show that CCM significantly improves pre-training efficiency. Specifically, the model trained with CCM shows comparative performance with the original BERT on the General Language Understanding Evaluation benchmark at half of the training cost.

Commonsense inference poses a unique challenge to reason and generate the physical, social, and causal conditions of a given event. Existing approaches to commonsense inference utilize commonsense transformers, which are large-scale language models that learn commonsense knowledge graphs. However, they suffer from a lack of coverage and expressive diversity of the graphs, resulting in a degradation of the representation quality. In this paper, we focus on addressing missing relations in commonsense knowledge graphs, and propose a novel contrastive learning framework called SOLAR. Our framework contrasts sets of semantically similar and dissimilar events, learning richer inferential knowledge compared to existing approaches. Empirical results demonstrate the efficacy of SOLAR in commonsense inference of diverse commonsense knowledge graphs. Specifically, SOLAR outperforms the state-of-the-art commonsense transformer on commonsense inference with ConceptNet by 1.84% on average among 8 automatic evaluation metrics. In-depth analysis of SOLAR sheds light on the effects of the missing relations utilized in learning commonsense knowledge graphs.

Word embedding is considered an essential factor in improving the performance of various Natural Language Processing (NLP) models. However, it is hardly applicable in real-world datasets as word embedding is generally studied with a well-refined corpus. Notably, in Hangeul (Korean writing system), which has a unique writing system, various kinds of Out-Of-Vocabulary (OOV) appear from typos. In this paper, we propose a robust Hangeul word embedding model against typos, while maintaining high performance. The proposed model utilizes a Convolutional Neural Network (CNN) architecture with a channel attention mechanism that learns to infer the original word embeddings. The model train with a dataset that consists of a mix of typos and correct words. To demonstrate the effectiveness of the proposed model, we conduct three kinds of intrinsic and extrinsic tasks. While the existing embedding models fail to maintain stable performance as the noise level increases, the proposed model shows stable performance.

Warning: This manuscript contains a certain level of offensive expression. As communication through social media platforms has grown immensely, the increasing prevalence of offensive language online has become a critical problem. Notably in Korea, one of the countries with the highest Internet usage, automatic detection of offensive expressions has recently been brought to attention. However, morphological richness and complex syntax of Korean causes difficulties in neural model training. Furthermore, most of previous studies mainly focus on the detection of abusive language, disregarding implicit offensiveness and underestimating a different degree of intensity. To tackle these problems, we present KOAS, a system that fully exploits both contextual and linguistic features and estimates an offensiveness score for a text. We carefully designed KOAS with a multi-task learning framework and constructed a Korean dataset for offensive analysis from various domains. Refer for a detailed demonstration.

Pre-trained word embeddings are widely used in various fields. However, the coverage of pre-trained word embeddings only includes words that appeared in corpora where pre-trained embeddings are learned. It means that the words which do not appear in training corpus are ignored in tasks, and it could lead to the limited performance of neural models. In this paper, we propose a simple yet effective method to represent out-of-vocabulary (OOV) words. Unlike prior works that solely utilize subword information or knowledge, our method makes use of both information to represent OOV words. To this end, we propose two stages of representation learning. In the first stage, we learn subword embeddings from the pre-trained word embeddings by using an additive composition function of subwords. In the second stage, we map the learned subwords into semantic networks (e.g., WordNet). We then re-train the subword embeddings by using lexical entries on semantic lexicons that could include newly observed subwords. This two-stage learning makes the coverage of words broaden to a great extent. The experimental results clearly show that our method provides consistent performance improvements over strong baselines that use subwords or lexical resources separately.

Distributed representations of words have been an indispensable component for natural language processing (NLP) tasks. However, the large memory footprint of word embeddings makes it challenging to deploy NLP models to memory-constrained devices (e.g., self-driving cars, mobile devices). In this paper, we propose a novel method to adaptively compress word embeddings. We fundamentally follow a code-book approach that represents words as discrete codes such as (8, 5, 2, 4). However, unlike prior works that assign the same length of codes to all words, we adaptively assign different lengths of codes to each word by learning downstream tasks. The proposed method works in two steps. First, each word directly learns to select its code length in an end-to-end manner by applying the Gumbel-softmax tricks. After selecting the code length, each word learns discrete codes through a neural network with a binary constraint. To showcase the general applicability of the proposed method, we evaluate the performance on four different downstream tasks. Comprehensive evaluation results clearly show that our method is effective and makes the highly compressed word embeddings without hurting the task accuracy. Moreover, we show that our model assigns word to each code-book by considering the significance of tasks.

Deep neural network-based pretraining methods have achieved impressive results in many natural language processing tasks including text classification. However, their applicability to large-scale text classification with numerous categories (e.g., several thousands) is yet to be well-studied, where the training data is insufficient and skewed in terms of categories. In addition, existing pretraining methods usually involve excessive computation and memory overheads. In this paper, we develop a novel multi-pretraining framework for large-scale text classification. This multi-pretraining framework includes both a self-supervised pretraining and a weakly supervised pretraining. We newly introduce an out-of-context words detection task on the unlabeled data as the self-supervised pretraining. It captures the topic-consistency of words used in sentences, which is proven to be useful for text classification. In addition, we propose a weakly supervised pretraining, where labels for text classification are obtained automatically from an existing approach. Experimental results clearly show that both pretraining approaches are effective for large-scale text classification task. The proposed scheme exhibits significant improvements as much as 3.8% in terms of macro-averaging F1-score over strong pretraining methods, while being computationally efficient.

In this paper, we present an adaptive convolution for text classification to give flexibility to convolutional neural networks (CNNs). Unlike traditional convolutions which utilize the same set of filters regardless of different inputs, the adaptive convolution employs adaptively generated convolutional filters conditioned on inputs. We achieve this by attaching filter-generating networks, which are carefully designed to generate input-specific filters, to convolution blocks in existing CNNs. We show the efficacy of our approach in existing CNNs based on the performance evaluation. Our evaluation indicates that all of our baselines achieve performance improvements with adaptive convolutions as much as up to 2.6 percentage point in seven benchmark text classification datasets.

Distributed representations of words play a major role in the field of natural language processing by encoding semantic and syntactic information of words. However, most existing works on learning word representations typically regard words as individual atomic units and thus are blind to subword information in words. This further gives rise to a difficulty in representing out-of-vocabulary (OOV) words. In this paper, we present a character-based word representation approach to deal with this limitation. The proposed model learns to generate word representations from characters. In our model, we employ a convolutional neural network and a highway network over characters to extract salient features effectively. Unlike previous models that learn word representations from a large corpus, we take a set of pre-trained word embeddings and generalize it to word entries, including OOV words. To demonstrate the efficacy of the proposed model, we perform both an intrinsic and an extrinsic task which are word similarity and language modeling, respectively. Experimental results show clearly that the proposed model significantly outperforms strong baseline models that regard words or their subwords as atomic units. For example, we achieve as much as 18.5% improvement on average in perplexity for morphologically rich languages compared to strong baselines in the language modeling task.