arXiv Science⌕ Search

arXiv subjects

Amila Weerasinghe

Publications and source records attributed to Amila Weerasinghe.

4 recordsLinked to original sources

Vocabulary-Free Next-Event Prediction on Temporal Graphs

Autoregressive transformer training through next-token prediction has driven remarkable advances in language modeling and has become increasingly effective for time-series modeling. Yet its success in temporal graph learning remains limited, and most temporal link-prediction methods are discriminative rather than autoregressive. While temporal graphs can have a natural serialization through chronological ordering edges, they lack an obvious graph-independent tokenization: existing node tokenization often uses a vocabulary that grows with the number of nodes, limiting scalability and cross-graph training. We propose GTGT, Generative Temporal Graph Transformer, a simple yet powerful vocabulary-free framework for self-supervised next-event prediction on temporal graphs. It predicts events for observed destinations, behavioral descriptors for novel destinations, and time gaps before new event. These targets have shared meanings across graphs, allowing one transformer to train across disjoint node sets without node-specific learned parameters. Local node identifiers let attention capture graph structure: Theoretically, we can show that our transformer-based model has the capacity for time-respecting message passing mechanisms. On Temporal Graph Benchmark (TGB), our GTGT not only achieves comparable or better link-prediction performance than established baselines, but also has much lower training cost. These results open the door to temporal graph pretraining for forecasting and other downstream tasks, providing a concrete step toward temporal graph foundation models.

cs.LG↗

ACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal Evidence

Large language models are increasingly used as scientific assistants, yet we still lack reliable ways to evaluate whether they can investigate a system rather than merely answer a prompt. A useful evaluation must let agents choose experiments, interpret finite evidence, and make causal decisions, while preserving objective ground truth. We introduce ACED (Active Causal Evidence and Decision Benchmark), a neuro-symbolic benchmark that separates the readable scientific surface from the formal causal world: natural-language scenarios describe the system, while a hidden causal model defines the data, the effect of every intervention, and the oracle answers. In ACED-Repair, agents diagnose a failing production system: within 15 experiments they must decide which controls to test, read confounded and filtered data, and recommend a fix, which we re-simulate on a hidden true objective. In ACED-Query, agents answer causal questions and make constrained causal decisions in worlds backed by categorical probabilistic graphical models. Acceptance on ACED-Repair ranges from 43% to 0% across 12 LLMs, while strong agents approach ceiling on ACED-Query's qualification questions. Tracing every experiment shows where investigations stop: most open-weight models test a true lever on its own in fewer than half of the worlds, the strongest models test the right levers but often miss the working dose or combination, and agents that do run a working fix almost always recommend it. No model reliably repairs systems with two independent causes or systems that need a subgroup-specific policy. These results suggest that reliable AI co-scientists need explicit support for choosing experiments, keeping track of evidence, and composing interventions. Code and data are available at https://github.com/Rose-STL-Lab/ACED-Bench.

cs.AI↗

Bounds for low-energy spectral properties of center-of-mass conserving positive two-body interactions

We study the low-energy spectral properties of positive center-of-mass conserving two-body Hamiltonians as they arise in models of fractional quantum Hall states. Starting from the observation that positive many-body Hamiltonians must have ground-state energies that increase monotonously in particle number, we explore what general additional constraints can be obtained for two-body interactions with "center-of-mass conservation" symmetry, both in the presence and absence of particle-hole symmetry. We find general bounds that constrain the evolution of the ground-state energy with particle number, and in particular, constrain the chemical potential at $T=0$. Special attention is given to Hamiltonians with zero modes, in which case similar bounds on the first excited state are also obtained, using a duality property. In this case, in particular, an upper bound on the charge gap is also obtained. We further comment on center of mass and relative decomposition in disk geometry within the framework of second quantization.

cond-mat.str-el↗

Thin torus perturbative analysis of elementary excitations in the Gaffnian and Haldane-Rezayi quantum Hall states

We present a systematic perturbative approach to study excitations in the thin cylinder/torus limit of the quantum Hall states. The approach is applied to the Haldane-Rezayi and Gaffnian quantum Hall states, which are both expected to have gapless excitations in the usual two-dimensional thermodynamic limit. For the Haldane-Rezayi state, we confirm that gapless excitations are present also in the "one-dimensional" thermodynamic limit of an infinite thin cylinder, in agreement with earlier considerations based on the wave functions alone. In contrast, we identify the lowest excitations of the Gaffnian state in the thin cylinder limit, and conclude that they are gapped, using a combination of perturbative and numerical means. We discuss possible scenarios for the cross-over between the two-dimensional and the one-dimensional thermodynamic limit in this case.

cond-mat.str-el↗