Search papers, labs, and topics across Lattice.
Email:
13
0
11
AI malfunctions, like Clever Hans effects, can be directly traced to specific artifact patterns in the input data using a model-agnostic dictionary of interpretable concepts.
Deep GNNs can achieve superior depth scaling and accuracy by leveraging controlled Turing instability or stable near-critical propagation to combat over-smoothing.
A novel two-stream architecture disentangles temporal context from spatial refinement in EEG source imaging, achieving superior localization accuracy without compromising inference speed.
LUCAID outperforms experienced pathologists by achieving 93% concordance in lung cancer diagnostics, setting a new standard for AI in precision oncology.
ReactionAtlas uncovers nearly 47,000 reactions from just eight seed molecules, revolutionizing our understanding of carbohydrate chemistry and its implications for the origins of life.
A new attribution method reveals the full spectrum of information flow in ETGNNs, enhancing explainability beyond traditional event-related embeddings.
Atlas H&E-TME achieves expert-level accuracy in tissue profiling, turning standard H&E slides into a powerful tool for cancer research.
Symb-xMIL transforms MIL interpretability by revealing the hidden logical rules that govern model predictions, exposing decision patterns that traditional heatmaps miss.
Pretraining on diverse synthetic data allows a single model to excel across multiple MIL tasks with minimal labeled input, outperforming conventional supervised approaches.
Exponential complexity in GNN explainability is no longer a barrier: this work achieves linear-time subgraph attribution via message passing.
Observed sample displacements can be integrated into optimal transport to carve expressways through the input space, leading to more reliable modeling of distribution shifts.
Human-inspired context sensitivity boosts visual reasoning in machines, closing the gap between AI and human perception.
Attention heatmaps in MIL models for histopathology are often misleading, and simpler methods like perturbation or LRP provide more faithful explanations.
