Piotr Koniusz

Associate Professor@University of New South Wales·澳大利亚·OpenReview

研究方向

node embeddings · node classification · graph classification · higher-order tensors · RKHS · bilinear pooling · feature learning · zero- · one- and few-shot learning · domain adaptation · self-supervised learning · contrastive learning · hyperbolic learning · dimensional collapse · LogDet divergence · adversarial learning · computer vision · VLM

Piotr Koniusz

Tug-of-War No More: Harmonizing Accuracy and Robustness in Vision-Language Models via Stability-Aware Task Vector Merging

Math Blind: Failures in Diagram Understanding Undermine Reasoning in MLLMs

DeCo-DETR: Decoupled Cognition DETR for efficient Open-Vocabulary Object Detection

C3-OWD: A Curriculum Cross-modal Contrastive Learning Framework for Open-World Detection

Subspace Kernel Learning on Tensor Sequences

Scalable Continuous-Time Hidden Markov Models

Efficient Fine-Tuning via Behavior-Guided Spectral Alignment

Towards Interference-Free One-Shot Federated Learning via Frequency Domain Aggregation

Provably Continual Unlearning for Large Language Model

When the Brain Sees Beyond Pixels: Creative Brain-to-Vision Reconstruction

Orthogonal Updates Are Optimal for Continual Learning

CodeBiasBench: Benchmarking social fairness of large language model generated code

Robust SuperAlignment: Weak-to-Strong Robustness Generalization for Vision-Language Models

CrossSpectra: Exploiting Cross-Layer Smoothness for Parameter-Efficient Fine-Tuning

Machine Unlearning via Task Simplex Arithmetic

Improving Zero-Shot Adversarial Robustness in Vision-Language Models by Closed-form Alignment of Adversarial Path Simplices

Graph Your Own Prompt

Amortized Active Generation of Pareto Sets

Primitive Vision: Improving Diagram Understanding in MLLMs

Less is More: Extreme Gradient Boost Rank-1 Adaption for Efficient Finetuning of LLMs

Learnable Expansion of Graph Operators for Multi-Modal Feature Fusion

Open Eyes, Then Reason: Fine-grained Visual Mathematical Understanding in MLLMs

Equivariant Graph Self-Attention Transformer for Learning Higher-Order Interactions in 3D Molecular Structures