Showing 1–7 of 7 results for author: Song, J Y

Exploring Practitioner Perspectives On Training Data Attribution Explanations

Authors: Elisa Nguyen, Evgenii Kortukov, Jean Y. Song, Seong Joon Oh

Abstract: Explainable AI (XAI) aims to provide insight into opaque model reasoning to humans and as such is an interdisciplinary field by nature. In this paper, we interviewed 10 practitioners to understand the possible usability of training data attribution (TDA) explanations and to explore the design space of such an approach. We confirmed that training data quality is often the most important factor for… ▽ More Explainable AI (XAI) aims to provide insight into opaque model reasoning to humans and as such is an interdisciplinary field by nature. In this paper, we interviewed 10 practitioners to understand the possible usability of training data attribution (TDA) explanations and to explore the design space of such an approach. We confirmed that training data quality is often the most important factor for high model performance in practice and model developers mainly rely on their own experience to curate data. End-users expect explanations to enhance their interaction with the model and do not necessarily prioritise but are open to training data as a means of explanation. Within our participants, we found that TDA explanations are not well-known and therefore not used. We urge the community to focus on the utility of TDA techniques from the human-machine collaboration perspective and broaden the TDA evaluation to reflect common use cases in practice. △ Less

Submitted 22 November, 2023; v1 submitted 31 October, 2023; originally announced October 2023.

Comments: Accepted to NeurIPS XAI in Action workshop 2023

Neglected Free Lunch -- Learning Image Classifiers Using Annotation Byproducts

Authors: Dongyoon Han, Junsuk Choe, Seonghyeok Chun, John Joon Young Chung, Minsuk Chang, Sangdoo Yun, Jean Y. Song, Seong Joon Oh

Abstract: Supervised learning of image classifiers distills human knowledge into a parametric model through pairs of images and corresponding labels (X,Y). We argue that this simple and widely used representation of human knowledge neglects rich auxiliary information from the annotation procedure, such as the time-series of mouse traces and clicks left after image selection. Our insight is that such annotat… ▽ More Supervised learning of image classifiers distills human knowledge into a parametric model through pairs of images and corresponding labels (X,Y). We argue that this simple and widely used representation of human knowledge neglects rich auxiliary information from the annotation procedure, such as the time-series of mouse traces and clicks left after image selection. Our insight is that such annotation byproducts Z provide approximate human attention that weakly guides the model to focus on the foreground cues, reducing spurious correlations and discouraging shortcut learning. To verify this, we create ImageNet-AB and COCO-AB. They are ImageNet and COCO training sets enriched with sample-wise annotation byproducts, collected by replicating the respective original annotation tasks. We refer to the new paradigm of training models with annotation byproducts as learning using annotation byproducts (LUAB). We show that a simple multitask loss for regressing Z together with Y already improves the generalisability and robustness of the learned models. Compared to the original supervised learning, LUAB does not require extra annotation costs. ImageNet-AB and COCO-AB are at https://github.com/naver-ai/NeglectedFreeLunch. △ Less

Submitted 26 July, 2023; v1 submitted 30 March, 2023; originally announced March 2023.

Comments: Code & data at https://github.com/naver-ai/NeglectedFreeLunch. To be presented at ICCV'23

ModSandbox: Facilitating Online Community Moderation Through Error Prediction and Improvement of Automated Rules

Authors: Jean Y. Song, Sangwook Lee, Jisoo Lee, Mina Kim, Juho Kim

Abstract: Despite the common use of rule-based tools for online content moderation, human moderators still spend a lot of time monitoring them to ensure that they work as intended. Based on surveys and interviews with Reddit moderators who use AutoModerator, we identified the main challenges in reducing false positives and false negatives of automated rules: not being able to estimate the actual effect of a… ▽ More Despite the common use of rule-based tools for online content moderation, human moderators still spend a lot of time monitoring them to ensure that they work as intended. Based on surveys and interviews with Reddit moderators who use AutoModerator, we identified the main challenges in reducing false positives and false negatives of automated rules: not being able to estimate the actual effect of a rule in advance and having difficulty figuring out how the rules should be updated. To address these issues, we built ModSandbox, a novel virtual sandbox system that detects possible false positives and false negatives of a rule to be improved and visualizes which part of the rule is causing issues. We conducted a user study with online content moderators, finding that ModSandbox can support quickly finding possible false positives and false negatives of automated rules and guide moderators to update those to reduce future errors. △ Less

Submitted 17 October, 2022; originally announced October 2022.

Comments: 24 pages, 10 figures

Magnetic evolution of Spinel Mn1-xZnxCr2O4 single crystals

Authors: G. T. Lin, X. Luo, Q. L. Pei, F. C. Chen, C. Yang, J. Y. Song, L. H. Yin, W. H. Song, Y. P. Sun

Abstract: Mn1-xZnxCr2O4 (0<x<1) single crystals have been grown by the chemical vapor transport (CVT) method. The crystallographic, magnetic, and thermal transport properties of the single crystals were investigated by the room-temperature X-ray diffraction, magnetization M(T) and specific heat CP(T) measurements. Mn1-xZnxCr2O4 crystals show a cubic structure, the lattice constant a decreases with the incre… ▽ More Mn1-xZnxCr2O4 (0<x<1) single crystals have been grown by the chemical vapor transport (CVT) method. The crystallographic, magnetic, and thermal transport properties of the single crystals were investigated by the room-temperature X-ray diffraction, magnetization M(T) and specific heat CP(T) measurements. Mn1-xZnxCr2O4 crystals show a cubic structure, the lattice constant a decreases with the increasing content x of the doped Zn2+ ions and follows the Vegard law. Based on the magnetization and heat capacity measurements, the magnetic evolution of Mn1-xZnxCr2O4 crystals has been discussed. For 0<x<0.3, the magnetic ground state is the coexistence of the collinear ferrimagnetic order (CFIM) and spiral ferrimagnetic one (SFIM), which is similar to that of the parent MnCr2O4. When x changes from 0.3 to 0.8, the SFIM is progressively suppressed and spin glass-like behavior is observed. When x is above 0.8, an antiferromagnetic (AFM) order presents. At the same time, the magnetic specific heat (Cmag.) was also investigated and the results are coincident with the magnetic measurements. The possible reasons based on the disorder effect and the reduced molecular field effect induced by the substitution of Mn2+ ions by nonmagnetic Zn2+ ones in Mn1-xZnxCr2O4 crystals have been discussed. △ Less

Submitted 25 February, 2016; originally announced February 2016.

Spin dynamics, electronic and thermal transport properties of two-dimensional CrPS4 single crystal

Authors: Q. L. Pei, X. Luo, G. T. Lin, J. Y. Song, L. Hu, Y. M. Zou, L. Yu, W. Tong, W. H. Song, W. J. Lu, Y. P. Sun

Abstract: 2-Dimensional (2D) CrPS4 single crystals have been grown by the chemical vapor transport method. The crystallographic, magnetic, electronic and thermal transport properties of the single crystals were investigated by the room-temperature X-ray diffraction, electrical resistivity \r{ho}(T), specific heat CP(T) and the electronic spin response (ESR) measurements. CrPS4 crystals crystallize into a mo… ▽ More 2-Dimensional (2D) CrPS4 single crystals have been grown by the chemical vapor transport method. The crystallographic, magnetic, electronic and thermal transport properties of the single crystals were investigated by the room-temperature X-ray diffraction, electrical resistivity \r{ho}(T), specific heat CP(T) and the electronic spin response (ESR) measurements. CrPS4 crystals crystallize into a monoclinic structure. The electrical resistivity \r{ho}(T) shows a semiconducting behavior with an energy gap Ea=0.166 eV. The antiferromagntic (AFM) transition temperature is about TN=36 K. The spin flip** induced by the applied magnetic field is observed along the c axis. The magnetic phase diagram of CrPS4 single crystal has been discussed. The extracted magnetic entropy at TN is about 10.8 J/mol K, which is consistent with the theoretical value R ln(2S + 1) for S = 3/2 of the Cr3+ ion. Based on the mean-field theory, the magnetic exchange constants J1 and Jc corresponding to the interactions of the intralayer and between layers are about 0.143 meV and -0.955 meV are obtained based on the fitting of the susceptibility above TN, which agree with the results obtained from the ESR measurements. With the help of the strain for tuning the magnetic properties, monolayer CrPS4 may be a promising candidate to explore 2D magnetic semiconductors. △ Less

Submitted 25 February, 2016; originally announced February 2016.

Journal ref: J. Appl. Phys. 119, 043902 (2016)

Superconductivity in CaSn3 single crystal with a AuCu3-type structure

Authors: X. Luo, D. F. Shao, Q. L. Pei, J. Y. Song, L. Hu, Y. Y. Han, X. B. Zhu, W. H. Song, W. J. Lu, Y. P. Sun

Abstract: We report the superconductivity of the CaSn3 single crystal with a AuCu3-type structure, namely cubic space group Pm3m. The superconducting transition temperature TC=4.2 K is determined by the magnetic susceptibility, electrical resistivity, and heat capacity measurements. The magnetization versus magnetic field (M-H) curve at low temperatures shows the typical-II superconducting behavior. The est… ▽ More We report the superconductivity of the CaSn3 single crystal with a AuCu3-type structure, namely cubic space group Pm3m. The superconducting transition temperature TC=4.2 K is determined by the magnetic susceptibility, electrical resistivity, and heat capacity measurements. The magnetization versus magnetic field (M-H) curve at low temperatures shows the typical-II superconducting behavior. The estimated lower and upper critical fields are about 125 Oe and 1.79 T, respectively. The penetration depth λ(0) and coherence length ξ(0) are calculated to be approximately 1147 nm and 136 nm by the Ginzburg-Landau equations. The estimated Sommerfeld coefficient of the normal state γ_N is about 2.9 mJ/mol K2. ΔC/γNTC =1.13 and λep=0.65 suggest that CaSn3 single crystal is a weakly coupled superconductor. Electronic band structure calculations show a complex multi-sheet Fermi surface formed by three bands and a low density of states (DOS) at the Fermi level, which is consistent with the experimental results. Based on the analysis of electron phonon coupling of AX3 compounds (A=Ca, La, and Y; X=Sn and Pb), we theoretically proposed a way to increase TC in the system. △ Less

Submitted 19 November, 2015; originally announced December 2015.

Journal ref: J. Mater. Chem. C,3, 11432 (2015)

Quantum electrical transport properties of topological insulator Bi2Te3 nanowires

Authors: Hong-Seok Kim, Ho Sun Shin, Joon Sung Lee, Chi Won Ahn, Jae Yong Song, Yong-Joo Doh

Abstract: We investigate the quantum transport properties of surface electrons on a topological insulator Bi2Te3 nanowire in a magnetotransport study. Although the nanowires are synthesized by using a relatively coarse method of electrochemical deposition, clear Aharonov-Bohm oscillations of phases 0 and pi are observed, owing to the highly coherent surface electron channel. The oscillation amplitude exhibi… ▽ More We investigate the quantum transport properties of surface electrons on a topological insulator Bi2Te3 nanowire in a magnetotransport study. Although the nanowires are synthesized by using a relatively coarse method of electrochemical deposition, clear Aharonov-Bohm oscillations of phases 0 and pi are observed, owing to the highly coherent surface electron channel. The oscillation amplitude exhibits exponential temperature dependence, suggesting that the phase coherence length L_phi is inversely proportional to the temperature, as in quasi-ballistic systems. In addition, a weak antilocalization analysis on the surface channel by using a one-dimensional localization theory, enabled by successful extraction of the surface contribution from the magnetoconductance data, is provided in support of the temperature dependence of L_phi. △ Less

Submitted 31 October, 2015; originally announced November 2015.

Journal ref: Current Applied Physics, vol.16 pg.51 (2016)