๐ข Call for Papers โ LoG 2026
Paper submissions are now open for LoG 2026!
Submit your work on any areas broadly related to learning on graphs and geometry
๐ https://t.co/OAVHVYCSeg
๐ Key dates (AoE):
โข July 29 โ Abstract deadline
โข Aug 1 โ Full paper deadline
#LoG2026
Can we evaluate if LLMs can reason structurally?
I discussed this question through the lens of data structures in a talk at @SimonsInstitute. In a joint ICML'26 work with @yuhe441, Yingxi Li, and @crwhite_ml, we introduce DSR-Bench, the data structure reasoning benchmark. (1/9)
[0/n] Can LLMs ๐ข๐ค๐ต๐ถ๐ข๐ญ๐ญ๐บ reason about structureโorder, hierarchy, connectivity, and how parts fit together?๐งฉ
We introduce ๐๐ฆ๐ฅ-๐๐ฒ๐ป๐ฐ๐ต: a data-structure benchmark designed to test this ๐๐ถ๐๐ต๐ผ๐๐ ๐๐ผ๐ผ๐น๐. Even SOTA LLMs still struggle in the hardest settings.
๐ข New position paper: โ๐ก๐ฒ๐๐ฟ๐ฎ๐น ๐๐น๐ด๐ผ๐ฟ๐ถ๐๐ต๐บ๐ถ๐ฐ ๐ฅ๐ฒ๐ฎ๐๐ผ๐ป๐ถ๐ป๐ด ๐ฅ๐ฒ๐พ๐๐ถ๐ฟ๐ฒ๐ ๐ฎ ๐๐น๐ฒ๐ฎ๐ฟ ๐ฆ๐ฐ๐ผ๐ฝ๐ฒ, ๐ง๐ต๐ฒ๐ผ๐ฟ๐ฒ๐๐ถ๐ฐ๐ฎ๐น ๐๐ผ๐๐ป๐ฑ๐ฎ๐๐ถ๐ผ๐ป๐, ๐ฎ๐ป๐ฑ ๐๐บ๐ฝ๐ถ๐ฟ๐ถ๐ฐ๐ฎ๐น ๐ฅ๐ถ๐ด๐ผ๐ฟโ
When and where should we use Neural Algorithmic Reasoning (NAR)? NAR is an exciting emerging area at the intersection of ๐๐ก๐๐จ๐จ๐๐๐๐ก ๐๐ก๐๐ค๐ง๐๐ฉ๐๐ข ๐๐๐จ๐๐๐ฃ and ๐ฃ๐๐ช๐ง๐๐ก ๐๐ค๐ข๐ฅ๐ช๐ฉ๐๐ฉ๐๐ค๐ฃ โ but its foundations are still underdeveloped.
In this paper, we argue that the field needs:
โจ a sharper ๐๐๐๐๐ฃ๐๐ฉ๐๐ค๐ฃ of what NAR is (and is not), including clearer distinctions from neighboring paradigms,
๐ stronger ๐ฉ๐๐๐ค๐ง๐ฎ for expressivity and generalization,
๐งช more rigorous ๐๐๐ฃ๐๐๐ข๐๐ง๐ ๐จ and connection to high-impact ๐๐ฅ๐ฅ๐ก๐๐๐๐ฉ๐๐ค๐ฃ๐จ.
[0/n] Can LLMs ๐ข๐ค๐ต๏ฟฝ๏ฟฝ๏ฟฝ๐ข๐ญ๐ญ๐บ reason about structureโorder, hierarchy, connectivity, and how parts fit together?๐งฉ
We introduce ๐๐ฆ๐ฅ-๐๐ฒ๐ป๐ฐ๐ต: a data-structure benchmark designed to test this ๐๐ถ๐๐ต๐ผ๐๐ ๐๐ผ๐ผ๐น๐. Even SOTA LLMs still struggle in the hardest settings.
[5/n] This is a joint work with my amazing collaborators Yingxi Li, @crwhite_ml, and Ellen Vitercik.
Paper: https://t.co/ium6fmz32p
Code: https://t.co/uBl55JXMbo
Data: https://t.co/40YUd3WNEG
The thing about a more focused, in-person conference is the spontaneous research discussions and friendships that felt natural and intimate. Grateful to be part of this community and to everyone who made LoG 2025 happen. Phoenix is great and will always hold a place in my heart.