arXiv subjects

Joonsuk Park

Publications and source records attributed to Joonsuk Park.

26 records · Page 2Linked to original sources

arXiv2023

KoSBi: A Dataset for Mitigating Social Bias Risks Towards Safer Large Language Model Application

Large language models (LLMs) learn not only natural text generation abilities but also social biases against different demographic groups from real-world data. This poses a critical risk when deploying LLM-based applications. Existing research and resources are not readily applicable in South Korea due to the differences in language and culture, both of which significantly affect the biases and targeted demographic groups. This limitation requires localized social bias datasets to ensure the safe and effective deployment of LLMs. To this end, we present KO SB I, a new social bias dataset of 34k pairs of contexts and sentences in Korean covering 72 demographic groups in 15 categories. We find that through filtering-based moderation, social biases in generated content can be reduced by 16.47%p on average for HyperCLOVA (30B and 82B), and GPT-3.

cs.CL↗

arXiv2023

SQuARe: A Large-Scale Dataset of Sensitive Questions and Acceptable Responses Created Through Human-Machine Collaboration

The potential social harms that large language models pose, such as generating offensive content and reinforcing biases, are steeply rising. Existing works focus on coping with this concern while interacting with ill-intentioned users, such as those who explicitly make hate speech or elicit harmful responses. However, discussions on sensitive issues can become toxic even if the users are well-intentioned. For safer models in such scenarios, we present the Sensitive Questions and Acceptable Response (SQuARe) dataset, a large-scale Korean dataset of 49k sensitive questions with 42k acceptable and 46k non-acceptable responses. The dataset was constructed leveraging HyperCLOVA in a human-in-the-loop manner based on real news headlines. Experiments show that acceptable response generation significantly improves for HyperCLOVA and GPT-3, demonstrating the efficacy of this dataset.

cs.CL↗

arXiv2022

Critic-Guided Decoding for Controlled Text Generation

Steering language generation towards objectives or away from undesired content has been a long-standing goal in utilizing language models (LM). Recent work has demonstrated reinforcement learning and weighted decoding as effective approaches to achieve a higher level of language control and quality with pros and cons. In this work, we propose a novel critic decoding method for controlled language generation (CriticControl) that combines the strengths of reinforcement learning and weighted decoding. Specifically, we adopt the actor-critic framework to train an LM-steering critic from non-differentiable reward models. And similar to weighted decoding, our method freezes the language model and manipulates the output token distribution using called critic, improving training efficiency and stability. Evaluation of our method on three controlled generation tasks, namely topic control, sentiment control, and detoxification, shows that our approach generates more coherent and well-controlled texts than previous methods. In addition, CriticControl demonstrates superior generalization ability in zero-shot settings. Human evaluation studies also corroborate our findings.

cs.CL↗

arXiv2022

Masked Summarization to Generate Factually Inconsistent Summaries for Improved Factual Consistency Checking

Despite the recent advances in abstractive summarization systems, it is still difficult to determine whether a generated summary is factual consistent with the source text. To this end, the latest approach is to train a factual consistency classifier on factually consistent and inconsistent summaries. Luckily, the former is readily available as reference summaries in existing summarization datasets. However, generating the latter remains a challenge, as they need to be factually inconsistent, yet closely relevant to the source text to be effective. In this paper, we propose to generate factually inconsistent summaries using source texts and reference summaries with key information masked. Experiments on seven benchmark datasets demonstrate that factual consistency classifiers trained on summaries generated using our method generally outperform existing models and show a competitive correlation with human judgments. We also analyze the characteristics of the summaries generated using our method. We will release the pre-trained model and the code at https://github.com/hwanheelee1993/MFMA.

cs.CL↗

arXiv2022

Plug-and-Play Adaptation for Continuously-updated QA

Language models (LMs) have shown great potential as implicit knowledge bases (KBs). And for their practical use, knowledge in LMs need to be updated periodically. However, existing tasks to assess LMs' efficacy as KBs do not adequately consider multiple large-scale updates. To this end, we first propose a novel task--Continuously-updated QA (CuQA)--in which multiple large-scale updates are made to LMs, and the performance is measured with respect to the success in adding and updating knowledge while retaining existing knowledge. We then present LMs with plug-in modules that effectively handle the updates. Experiments conducted on zsRE QA and NQ datasets show that our method outperforms existing approaches. We find that our method is 4x more effective in terms of updates/forgets ratio, compared to a fine-tuning baseline.

cs.CL↗

arXiv2018

Black Arsenic: A Layered Semiconductor with Extreme in-plane Anisotropy

Two-dimensional (2D) layered materials emerge in recent years as a new platform to host novel electronic, optical or excitonic physics and develop unprecedented nanoelectronic and energy applications. By definition, these materials are strongly anisotropic between within the basal plane and cross the plane. The structural and property anisotropies inside their basal plane, however, are much less investigated. Herein, we report a rare chemical form of arsenic, called black-arsenic (b-As), as an extremely anisotropic layered semiconductor. We have performed systematic characterization on the structural, electronic, thermal and electrical properties of b-As single crystals, with particular focus on its anisotropies along two in-plane principle axes, armchair (AC) and zigzag (ZZ). Our analysis shows that b-As exhibits higher or comparable electronic, thermal and electric transport anisotropies between the AC and ZZ directions than any other known 2D crystals. Such extreme in-plane anisotropies are able to potentially implement novel ideas for scientific research and device applications.

cond-mat.mtrl-sci↗

arXiv2017

Argument Mining with Structured SVMs and RNNs

We propose a novel factor graph model for argument mining, designed for settings in which the argumentative relations in a document do not necessarily form a tree structure. (This is the case in over 20% of the web comments dataset we release.) Our model jointly learns elementary unit type classification and argumentative relation prediction. Moreover, our model supports SVM and RNN parametrizations, can enforce structure constraints (e.g., transitivity), and can express dependencies between adjacent relations and propositions. Our approaches outperform unstructured baselines in both web comments and argumentative essay datasets.

cs.CL↗

arXiv2017

Enhanced thermal conduction through nanostructured interfaces

Interfaces dominate heat conduction in nanostructured systems, and much work has focused on methods to enhance interfacial conduction. These approaches generally address planar interfaces, where the heat flux vector is everywhere normal to the interface. Here, we explore a nanostructured interface geometry that uses nonplanar features to enhance the effective interfacial conductance beyond what is possible with planar interfaces. This interface consists of interdigitating Al pillars embedded within SiO2 with characteristic feature size ranging from 100 nm to 800 nm. The total sidewall surface area is modulated to highlight the impact of this additional channel by changing the pillar-to-pillar pitch L_P between 1.6 um and 200 nm while maintaining the same Al:SiO2 fill fraction. Using optical pump-probe thermoreflectance measurements, we show that the effective conductance of a ~65 nm thick fin layer monotonically increases with decreasing L_P, and that the conductance for L_P = 200 nm is more than twice the prediction for a layered stack with the same volume ratio and a planar interface. Through a combination of Boltzmann transport modeling and finite element calculations, we explore the impact of the pitch L_P and the pillar aspect ratio on effective thermal conductance. This analysis suggests that the concept of nanostructured interfaces can be extended to interfaces between diffusive and quasi-ballistic media in highly scaled devices. Our work proposes that the controlled texturing of interfaces can facilitate interfacial conduction beyond the planar interface regime, opening new avenues for thermal management at the nanoscale.

cond-mat.mes-hall↗