Blätter-Navigation

An­ge­bot 51 von 90 vom 27.09.2019, 11:54

logo

Cha­rité - Uni­ver­si­täts­me­di­zin Ber­lin - CC15 Neur­o­logy, Bern­stein Cen­ter for Com­pu­ta­tional Neur­os­cience

The Char­ité - Uni­versitätsmed­izin Ber­lin is a joint med­ical fac­ulty, which serves both Freie Uni­versität Ber­lin and Hum­boldt Uni­versität zu Ber­lin. As one of the largest uni­versity hos­pit­als in Europe with an import­ant his­tory, it plays a lead­ing role in research, teach­ing and clin­ical care. The Char­ité uni­versity hos­pital has also made a name for itself as a mod­ern busi­ness with cer­ti­fic­a­tions in the med­ical, clin­ical and man­age­ment fields

Research Assist­ant (PhD stu­dent)

Research assist­ant pos­i­tion with option to pur­sue a PhD in machine learn­ing and inter­pretab­il­ity within ERC-fun­ded pro­ject

Work­ing field:

The topic of this PhD pro­ject is the the­ory and prac­tice of inter­pret­ing machine learn­ing mod­els.

State-of-the-art ML sys­tems are “black boxes” whose internal work­ings are too com­plex to be com­pre­hens­ible by a human. Recently, sev­eral explan­a­tion/inter­pret­a­tion frame­works for neural net­works and related tech­niques have emerged. But it has also been shown that these inter­pret­a­tions can be mis­lead­ing. In this PhD pro­ject, the suc­cess­ful applic­ant will develop novel meth­ods for explain­ing ML mod­els. The work com­prises, among other tasks,

  • to con­ceive math­em­at­ical defin­i­tions of inter­pretab­il­ity.
  • to gen­er­ate syn­thetic data with known inter­pret­a­tion, and to use them to bench­mark exist­ing explan­a­tion meth­ods.
  • to develop new meth­ods to inter­pret non-lin­ear machine learn­ing mod­els (deep neural net­works, ker­nel meth­ods).
  • to pub­lish the developed meth­ods as user-friendly open-source tool­boxes writ­ten in Mat­lab/Python.
  • to apply inter­pretable machine learn­ing mod­els to rel­ev­ant prob­lems in the health domain.
  • to con­duct lit­er­at­ure sur­veys, co-organ­ize work­shops.
  • to pub­lish research res­ults in rel­ev­ant sci­entific journ­als and present res­ults as talks/posters at rel­ev­ant con­fer­ences.

See braindata.char­ite.de for fur­ther inform­a­tion on the pos­i­tion and research group.

Require­ments:

  • Very good dip­loma, MSc, or equi­val­ent degree in a tech­nical field (e.g. machine learn­ing, com­puter sci­ence, stat­ist­ics, math­em­at­ics, com­pu­ta­tional (neuro) sci­ence, data sci­ence, phys­ics, elec­trical/bio­med­ical engin­eer­ing)
  • Strong back­ground in math­em­at­ics/stat­ist­ics and machine learn­ing
  • Good cod­ing skills (e.g., Mat­lab, Python, C++, Java)
  • Very good com­mand of writ­ten Eng­lish

How to ap­ply:

Applic­a­tions should include a let­ter of motiv­a­tion, a CV, tran­scripts and degree cer­ti­fic­ates, as well as (if avail­able) ref­er­ences, an Eng­lish-lan­guage writ­ing sample, and a cod­ing sample (e.g. link to a git­hub pro­ject).

Applic­a­tions should be sent by email to stefan.haufe@charite.de quot­ing the ref­er­ence num­ber. All doc­u­ments should be con­tained in a single pdf.

DIE CHAR­ITÉ – UNI­VERSITÄTSMED­IZIN BER­LIN makes its human resources decisions based on suit­ab­il­ity, com­pet­ence and pro­fes­sional per­form­ance. At the same time, it strives to increase the per­cent­age of women in man­age­ment pos­i­tions and takes this into con­sid­er­a­tion where can­did­ates are equally qual­i­fied within the lim­its of what is leg­ally pos­sible. Applic­a­tions from people with a migrant back­ground are also expli­citly wel­come. Severely dis­abled applic­ants are given pref­er­en­tial con­sid­er­a­tion in the case of can­did­ates with equal qual­i­fic­a­tions. A cer­ti­fic­ate of con­duct, or an exten­ded cer­ti­fic­ate of con­duct, if applic­able, must be sub­mit­ted. Any travel expenses incurred can­not be reim­bursed. Data pro­tec­tion notice: Char­ité points out that per­sonal data is stored and pro­cessed as part of the applic­a­tion pro­cess in dif­fer­ent areas of Char­ité (e.g. fac­ulty, staff com­mit­tee, human resources depart­ment). Fur­ther­more, the data may be trans­ferred or pro­cessed within the group as well as in loc­a­tions out­side the group (e.g. author­it­ies) to pro­tect legit­im­ate interests.