Proceeding of he 1 In erna ional Work hop on E plainable and In erpre able Machine Learning (XI-ML) Proceeding of he 1 In erna ional Work hop on E plainable and In erpre able Machine Learning (XI-ML) F8D65AC49921735A29615CA5879369FF GROBID - A machine learning software for extracting information from scholarly documents

Recen l , cien ific di co r e in ar ificial in elligence and da a cience ha foc ed on e plainable AI (XAI) i h re pec o algori hmic ran parenc , in erpre abili , acco n abili and finall e plainabili of algori hmic model and deci ion . In machine learning, approache can be cla ified a hi e-bo and black-bo . Whi e-bo approache , ch a r le learner and ind c i e programming, re l in e plici model hich are inheren l in erpre able (R din, 2019). On he o her hand, black-bo approache , ch a (deep) ne ral ne ork , re l in opaq e model . For hi econd pe of model , o er he la ear , differen approache for e -po e plana ion genera ion ha e been propo ed.

In hi ork hop, e an o bring oge her re earch from in erpre able and e plana or machine learning. In erpre able ML can profi from recen l propo ed e plana ion genera ion echniq e o make comple learned model more comprehen ible, e peciall o end-er (M ggle on, Schmid e al., 2018;F rnkran , Kliegr, Pa lheim, 2020;Lonjarre e al., 2020). In par ic lar, in erpre able learning can be in egra ed in o he con r c ion of comple model , e.g., for g iding heir con r c ion (A m eller e al., 2017), a ell a o refine he re pec i e model (Weidner, A m eller, Seipel, 2019). F r hermore, i can pro ide rich r le-ba ed echniq e o genera e in erpre able rroga e model for black-bo learner (Schmid, 2018). S ch rroga e model can be global model genera ed b r le-e rac ion mechani m (Haile ila ie, 2016) or local model hich allo richer local e plana ion han imple linear r le a , for in ance, propo ed b LIME (Rabold e al., 2019). Al o, a fron ier direc ion i in e iga ing p chological phenomena ha can affec he nder anding of machine learning model , ch a cogni i e bia e and con er a ional ma im (Kliegr, Bahnik, F rnkran , 2018). Thi in erdi ciplinar in pira ion, ch a debia ing echniq e long died b p chologi , ill hopef ll con rib e o a be er comprehen ibili of he re l of model crea ed b he ne genera ion of machine learning algori hm . XI-ML (E plainable and In erpre able Machine Learning) aim a bringing oge her re earch from in erpre able and e plainable machine learning. Hopef ll , in egra ing bo h area , allo ne per pec i e on q e ion on appropria e learning formali m , in erpre a ion and e plana ion echniq e , heir me ric , a ell a he re pec i e a e men op ion ari e.

The fir edi ion of he XI-ML (E plainable and In erpre able Machine Learning) ork hop a held on Sep ember 21, 2020 a he 43rd German Conference on Ar ificial In elligence, Bamberg, German . The ork hop a de o ed o he di c ion of he opic men ioned abo e. I aimed o pro ide an in erdi ciplinar for m o in e iga e f ndamen al i e in e plainable and in erpre able machine learning a ell a o di c recen ad ance , rend , and challenge in he e area . From 8 bmi ion (6 f ll and o hor paper ), 5 f ll paper and he o hor paper ere accep ed for pre en a ion in a comprehen i e re ie proce .

The remaining par of he ol me pre en re i ed er ion of paper ha ere di c ed d ring he ork hop. Bo r m e al. di c in heir paper abo ho o e plain m l i aria e ime erie foreca ing, in an applica ion o predic ing he S edi h GDP. Ne , M cha e al. pre en a po i ion paper on ho o con r c par icipa or de ign pace for he con e of e plainable AI in erface in e per domain . Af er ha , Volker di c ed ho he applica ion of he TED (Teaching E plana ion for Deci ion ) e plainable AI frame ork and he impac of cla (im-)balance. Flei , B ck and Thalmann pre en a hor paper on empirical re l in he con e of recr i ing -abo e plainabili and he in en ion o e AI-ba ed con er a ional agen . Po ka addre e fo nda ional i e o ard ol ing cla ifica ion problem i h q an i a i e ab rac arg men a ion. Mollenha er and A m eller pre en an approach for eq en ial e cep ional pa ern di co er ing pa ern-gro h (SEPP) -a he ba i of an e en ible frame ork for in erpre able machine learning on eq en ial da a. S n, Chakrabor i and Noble di c re l of a compara i e d of e plainer mod le in he con e of a oma ed kin le ion cla ifica ion. Finall , Marcin P. Joachimiak (En ironmen al Genomic and S em Biolog Di i ion, La rence Berkele Labora or ) kindl agreed o pre en a ke no e en i led Ho o each a comp er o learn abo microbe i h KG-COVID-19 . Thi alk in rod ced a ne re o rce ha amalgama e SARS-CoV-2 rela ed biological kno ledge from m l iple peciali ed kno ledge graph and on ologie . Wi h o er 10 million node , i i one of he large (if no he large ) re o rce of hi kind. In hi alk, Dr. Joachimiak demon ra ed he ili of hi re o rce for machine learning, empha i ing he need for e plainable echniq e .

Reference
Am eller, M., Ha a , N.,Schmid , A., & Kl pper, B. (2017). E plana ion-a are fea re elec ion ing mbolic ime erie ab rac ion: approache and e perience in a pe ro-chemical prod c ion con e . In IEEE I ). IEEE, Bo on, MA, USA F rnkran , J., Kliegr, T., & Pa lheim, H. (2020). On cogni i e preference and he pla ibili of r le-ba ed model . Mac L a , 109 (4), 853-898. Haile ila ie, T. (2016). R le e rac ion algori hm for deep ne ral ne ork : A re ie . a X a X :1610.05267 . Kliegr, Tom , p n Bahn k, and Johanne F rnkran . "A re ie of po ible effec of cogni i e bia e on in erpre a ion of r le-ba ed machine learning model ." arXi preprin arXi :1804.02969 (2018). Lonjarre , C., Robarde , C., Plan e i , M., A b r in, R., & A m eller, M. (2020). Wh Sho ld I Tr Thi I em? E plaining he Recommenda ion of an Model. In IEEE I a a C c Da a Sc c a d A a c . IEEE, Bo on, MA, USA M ggle on, S. H., Schmid, U., Zeller, C., Tamaddoni-Ne had, A., & Be old, T. (2018
•Mar in A m eller, O nabr ck Uni er i , German • Tom Kliegr, Uni er i of Economic Prag e, C ech Rep blic • U e Schmid, Uni er i of Bamberg, German Program Commi ee of XI-ML 2020 • Kla -Die er Al hoff, Uni er i of Hilde heim • Maria Bieliko a, Kem ele I i e f I ellige Tech l gie , Sl akia • Henrik Bo r m, KTH Ro al In i e of Technolog , S eden • Ami Dh randhar, IBM TJ Wa on Re earch Cen er, USA • Johanne F rnkran , Johanne Kepler Uni er i , Lin • Mar in Holena, C ech Academ of Science • E ke H llermeier, Uni er i of Paderborn • Kri ian Ker ing, TU Darm ad , German • Gr egor Nalepa, Jagellonian Uni er i , Poland • M kola Pecheni k i, TU Eindho en • Marc Plan e i , Uni er i L on • Eric Po ma, Tilb rg Uni er i • Celine Ro eirol, Uni er i Sorbonne Pari Nord • S efano Te o, KU Le en, Belgi m
Copyright (C) 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).