University of Ljubljana

Letting a cloud model write the program while a small on-device model answers bounded questions about user data that appear only at run time improved a code-only agent's success rate on AppWorld in all six backbone-model and test-partition combinations, by up to 14.9 percentage points, with user data staying on the device.

Letting a cloud model write the program while a small on-device model answers bounded questions about user data that appear only at run time improved a code-only agent's success rate on AppWorld in all six backbone-model and test-partition combinations, by up to 14.9 percentage points, with user data staying on the device.

Federated clustering previously revealed more model information to the server each round, and even without raw samples, repeated exchanges made client contributions linkable. eFDPC combines evolving fuzzy clustering with density peak clustering in a federated setting: each client evolves micro-clusters from private mini-batches and transmits only density-peak candidates as additive shares through a shuffler layer, so the server reconstructs peaks after mixing and selects global peaks. On synthetic and real benchmarks, it outperforms recent federated clustering baselines in one-shot settings and scales to higher-dimensional, non-IID data.

Federated clustering previously revealed more model information to the server each round, and even without raw samples, repeated exchanges made client contributions linkable. eFDPC combines evolving fuzzy clustering with density peak clustering in a federated setting: each client evolves micro-clusters from private mini-batches and transmits only density-peak candidates as additive shares through a shuffler layer, so the server reconstructs peaks after mixing and selects global peaks. On synthetic and real benchmarks, it outperforms recent federated clustering baselines in one-shot settings and scales to higher-dimensional, non-IID data.

The authors introduce the Graph Transformer Language Model (GTLM), which lets a pretrained LLM process graph topology natively by injecting graph-aware attention biases directly into its attention modules, adding only about 0.015% structure-related parameters; they prove the bidirectional attention prefix is permutation-equivariant over nodes and that GTLM reduces exactly to the pretrained model when no graph is present, and report that it matches or exceeds domain-specific state-of-the-art models on text-attributed graph benchmarks, GraphRAG on WebQSP, and molecular benchmarks, meaningfully improves over strong baselines on GraphQA, and keeps needle-in-a-graph accuracy flat from 1k to 64k tokens and 4x past the training length while an identically trained flat-text baseline collapses.

The authors introduce the Graph Transformer Language Model (GTLM), which lets a pretrained LLM process graph topology natively by injecting graph-aware attention biases directly into its attention modules, adding only about 0.015% structure-related parameters; they prove the bidirectional attention prefix is permutation-equivariant over nodes and that GTLM reduces exactly to the pretrained model when no graph is present, and report that it matches or exceeds domain-specific state-of-the-art models on text-attributed graph benchmarks, GraphRAG on WebQSP, and molecular benchmarks, meaningfully improves over strong baselines on GraphQA, and keeps needle-in-a-graph accuracy flat from 1k to 64k tokens and 4x past the training length while an identically trained flat-text baseline collapses.