
PhD Candidate @UKPLab | interested in representation learning, RAG & LLM Agents | MSc. AI from @MaastrichtU | prev. @DFKI #NLProc Supervisor: @Nils_Reimers
Maastricht
Joined June 2016
- Tweets73
- Following365
- Followers104
- Likes500
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
We are delighted to share that @IGurevych , the Director of the @UKPLab , has been recognized as one of theย ๐ง๐ผ๐ฝ ๐ญ๐ฌ๐ฌ ๐บ๐ผ๐๐ ๐ถ๐ป๐ณ๐น๐๐ฒ๐ป๐๐ถ๐ฎ๐น ๐๐ผ๐บ๐ฒ๐ป ๐ถ๐ป ๐๐ต๐ฒ ๐๐ฒ๐ฟ๐บ๐ฎ๐ป ๐ฒ๐ฐ๐ผ๐ป๐ผ๐บ๐ย by @manager_magazin and @BCG !ย ๐
(1/๐งต)
Im Krisenjahr 2024 sind noch mehr Frauen in Spitzenpositionen der deutschen Wirtschaft vorgerรผckt. Die wichtigsten Topmanagerinnen, Unternehmerinnen, Aufsichtsrรคtinnen und Wegbereiterinnen fรผr den Wandel vereint diese Liste. manager-magazin.de/unternehmโฆ
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
The Milner Award and Lecture 2025 is awarded to Professor Iryna Gurevych for her major contributions to natural language processing (NLP) and artificial intelligence that combine deep understanding of human language and cognitive faculty with the latest paradigms in machine learning. #RSMedals
royalsociety.org/medals-and-โฆ
Picture Credit: TU Darmstadt; Rรผdiger Dunker
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
Eine besondere Auszeichnung: Die TU-Informatikerin Iryna Gurevych @IGurevych erhรคlt als erste Wissenschaftlerin in Deutschland und als erste Universitรคtsprofessorin den renommierten โMilner Awardโ der britischen @royalsociety . Herzlichen Glรผckwunsch! tu-darmstadt.de/universitaetโฆ
The @RoyalSociety has just announced that @IGurevych will receive the 2025 #MilnerAward! Our director is being honored for contributions to #NLProc & #AI, combining an understanding of human cognitive abilities with the latest Machine Learning paradigms.
royalsociety.org/medals-and-โฆ
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
Meet the man who can tell you everything about Triple Encoders: Our colleague @J_st_s who presented a paper co-authored with @_florianmai (@CW_KULeuven), @Nils_Reimers (@cohere), @Gerasimoss (@MaastrichtU) and @IGurevych (@UKPLab) at #ACL2024NLP ๐น๐ญ!
nitter.cf/UKPLab/status/18229951โฆ
Modular Sentence Representations Are All You Need!
Idea: encode distances between pairwise utterances and keep the context!
Come with us on a journey of weightless sequential distributed composition! ๐งต (1/8) #ACL2024NLP
Poster: Tue 10:30
๐ฐ arxiv.org/abs/2402.12332
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
Happening now till 1200 at poster 15! @J_st_s is eagerly waiting #acl2024nlp
Replying to @gerasimoss
#acl2024nlp main: @J_st_s master thesis work @UM_DACS+@UKPLab: Triple-encoders improve contextualization & efficiency without re-encoding, achieving better zero-shot generalization & performance compared to bi-encoders๐ค๐ก
๐: arxiv.org/abs/2402.12332
nitter.cf/UKPLab/status/17950163โฆ
>>
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
Consider following the authors @J_st_s (@UKPLab), @_florianmai (@CW_KULeuven), @Nils_Reimers (@cohere), @Gerasimoss Spanakis (@MaastrichtU) and @IGurevych for more information or an exchange of ideas. (8/๐งต) See you this week at #ACL2024NLP in Thailand ๐น๐ญ!
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
Meet our fellow researchers representing @UKPLab at this year's @ACLmeeting: @IGurevych, @QianRuan_, @J_st_s, @androneil54, @trumancfy, Sheng Lu, @haishuofang, @Furkansahinuc, @KexinWang2049, @LHaausing, @AndreasWaldis, and a very special guest from @BathNLP, @Harish! #ACL2024NLP
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
#acl2024nlp main: @J_st_s master thesis work @UM_DACS+@UKPLab: Triple-encoders improve contextualization & efficiency without re-encoding, achieving better zero-shot generalization & performance compared to bi-encoders๐ค๐ก
๐: arxiv.org/abs/2402.12332
nitter.cf/UKPLab/status/17950163โฆ
>>
Replying to @UKPLab @AndreasWaldis @hslu @yufanghou @IBMResearch @IGurevych @imbugene @XiaodanZhu2048 @QueensECE @IngenuityLabs @gg42554 @Uni_WUE @M_Jonibek @SuJinyan6 @ArtemShelmanov @aktsvigun @OAfzal2 @AlhamFikri @nyhabash @preslav_nakov @mbzuai @TUDarmstadt @NYUAbuDhabi @Furkansahinuc @ilokuznetsov @IBigoulaeva @rachneet4 @harish @BathNLP
ยปTriple-Encoders: Representations That Fire Together, Wire Togetherยซ by @J_st_s (@UKPLab), @_florianmai (@CW_KULeuven), @Nils_Reimers (@cohere), @Gerasimoss Spanakis (@MaastrichtU) and @IGurevych (9/๐งต) #ACL2024NLP
๐ arxiv.org/abs/2402.12332
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
Replying to @UKPLab @AndreasWaldis @hslu @yufanghou @IBMResearch @IGurevych @imbugene @XiaodanZhu2048 @QueensECE @IngenuityLabs @gg42554 @Uni_WUE @M_Jonibek @SuJinyan6 @ArtemShelmanov @aktsvigun @OAfzal2 @AlhamFikri @nyhabash @preslav_nakov @mbzuai @TUDarmstadt @NYUAbuDhabi @Furkansahinuc @ilokuznetsov @IBigoulaeva @rachneet4 @harish @BathNLP
ยปTriple-Encoders: Representations That Fire Together, Wire Togetherยซ by @J_st_s (@UKPLab), @_florianmai (@CW_KULeuven), @Nils_Reimers (@cohere), @Gerasimoss Spanakis (@MaastrichtU) and @IGurevych (9/๐งต) #ACL2024NLP
๐ arxiv.org/abs/2402.12332
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
We are happy to announce that 11 papers authored or co-authored by UKP members have been accepted for publication at this year's @aclmeeting in Bangkok ๐น๐ญ. Congratulations to everyone involved! ๐ #ACL2024NLP
๐ informatik.tu-darmstadt.de/uโฆ
The accepted papers are (1/๐งต):
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
Attention #NLProc community! ๐ข Join the Perspective Argument Retrieval Shared Task, co-located with #argmining_2024 & #ACL2024NLP. Dive into the challenge of retrieving arguments tailored to diverse demographic profiles! ๐จ
๐ blubberli.github.io/perspectโฆ
๐ซ perspectivear-zzl7474.slack.โฆ
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
๐จ Sentence Transformers v2.3.0 has just been released!
Sentence Transformers is a framework developed at @UKPLab that allows an easy fine-tuning of custom embeddings models, to achieve maximal performance on your specific task.
Release notes: github.com/UKPLab/sentence-tโฆ
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
Auto-Regressive LLMs have a role to play: turning abstract ideas into token sequences (words, actions, code...).
But abstract ideas should be elaborated through planning/reasoning in representation space.
AR-LLMs go directly from prompt to answer, skipping the step of reasoning in representation space altogether.
i'm skeptical of @ylecun's JEPA models fully deprecating autoregressive LLMs. but he's just been right time and again; with deep neural nets first, parallel sequence models (instead of recurrent), the cake and unsupervised learning (over RL). so betting against him is a bad idea.
Many thanks to @_florianmai for an amazing talk at the @UKPLab on Attention alternatives.
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
intelligence is an emergent property of physics
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
For the die-hard-still-hungry-for-NLP @J_st_s is ready to pitch you our work on curved contrastive learning for sequence modelling (short- and long-term dialogue planning). โImagination is all you needโ #CODI2023 #ACL2023NLP #NLProc
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
Hey #ACL2023NLP peeps, I will be in Toronto ๐จ๐ฆ open to chat & also presenting
1๏ธโฃ Vendorlink (f/@vageeshsaxena) โฐ12/Jul 11-1230 nitter.cf/gerasimoss/status/1654โฆ
2๏ธโฃ Imagination is all you need! (f/@J_st_s): โฐ10/Jul 19-21, 11/Jul 11-1230, 14/Jul 1520-17 (CODI)
nitter.cf/gerasimoss/status/1654โฆ
Replying to @gerasimoss
The second paper is the work of @J_st_s (w/ Stefan Schaffer) on Curved Contrastive Learning: We "imagine" the likelihood of future patterns in multi-turn dialogues, specifically by ordering future goal utterances that are multiple turns away, given a dialogue context >> 4/5
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
๐ค Happy summer break from the #MNLP reading group! We will reconvene in September! @ThalesBertaglia @J_st_s @Khartraxx @antoiloui @HaoyangGui60578 @CatalinaGoanta
Justus-Jonas Erker ๐ช๐บ @ EACL 2026 ๐ฒ๐ฆ retweeted
Another lovely meeting of the Maastricht NLP Reading Group (master students, PhDs and UDs). This time we discussed MetaFormer (from Computer Vision) and the simple but powerful pooling i/o self-attention @J_st_s @abdrahman_issam @UM_DACS
























