Sepp Hochreiter

Minimax-Optimal Aggregation for Density Ratio Estimation

Rethinking Uncertainty Estimation in LLMs: A Principled Single-Sequence Measure

Addressing Pitfalls in the Evaluation of Uncertainty Estimation Methods for Natural Language Generation

xLSTM Scaling Laws: Competitive Performance with Linear Time-Complexity

AP-OOD: Attention Pooling for Out-of- Distribution Detection

TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning

Tiled Flash Linear Attention: More Efficient Linear RNN and xLSTM Kernels

Rethinking Losses for Diffusion Bridge Samplers

pLSTM: parallelizable Linear Source Transition Mark networks

xLSTM 7B: A Recurrent LLM for Fast and Efficient Inference

FlashRNN: I/O-Aware Optimization of Traditional RNNs on modern hardware

MIM-Refiner: A Contrastive Learning Boost from Intermediate Pre-Trained Masked Image Modeling Representations

The Disparate Benefits of Deep Ensembles

A Large Recurrent Action Model: xLSTM enables Fast Inference for Robotics Tasks

Improving Uncertainty Estimation through Semantically Diverse Language Generation

Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics

Parameter Efficient Fine-tuning via Explained Variance Adaptation

A Large Recurrent Action Model: xLSTM enables Fast Inference for Robotics Tasks

Bio-xLSTM: Generative modeling, representation and in-context learning of biological and chemical sequences

Vision-LSTM: xLSTM as Generic Vision Backbone

VN-EGNN: E(3)- and SE(3)-Equivariant Graph Neural Networks with Virtual Nodes Enhance Protein Binding Site Identification

Geometry-Informed Neural Networks

Geometry-Informed Neural Networks

One Initialization to Rule them All: Fine-tuning via Explained Variance Adaptation

Retrieval-Augmented Decision Transformer: External Memory for In-context RL

Large Language Models Can Self-Improve At Web Agent Tasks

The Disparate Benefits of Deep Ensembles

Rethinking Uncertainty Estimation in Natural Language Generation

Comparison Visual Instruction Tuning

On Information-Theoretic Measures of Predictive Uncertainty