Using Transformers for one-shot visual imitation, achieving ~2x task success rate improvement.

Sudeep Dasari, Abhinav Gupta

Introduced LRA benchmark for evaluating long-sequence Transformer models, improving efficiency.

Yi Tay, Mostafa Dehghani, Samira Abnar et al.

POMO exploits multiple starting points and symmetry in RL to solve TSP, CVRP, KP with 0.14% optimality gap, over tenfold inference speedup.

Yeong-Dae Kwon, Jinho Choo, Byoungjip Kim et al.

Proposes variational Bayesian unlearning using KL divergence minimization, applied to sparse Gaussian process and logistic regression, ensuring efficient data removal.

Quoc Phong Nguyen, Bryan Kian Hsiang Low, Patrick Jaillet

PEP constructs an ensemble of perturbed parameters via Gaussian noise, enhancing calibration and likelihood on pretrained models without additional training.

Alireza Mehrtash, Purang Abolmaesumi, Polina Golland et al.

Proposes Unlearning and Amnesiac Unlearning to efficiently remove sensitive data from trained neural networks, ensuring privacy and compliance.

Laura Graves, Vineel Nagisetty, Vijay Ganesh

Proposes a multi-agent DQN framework with K-means clustering for joint UAV trajectory and power optimization, boosting throughput in NOMA UAV networks.

Ruikang Zhong, Xiao Liu, Yuanwei Liu et al.

Introduces approximate information state framework for partial observation RL, enabling data-driven dynamic programming with performance guarantees.

Jayakumar Subramanian, Amit Sinha, Raihan Seraj et al.

RELIC enforces invariance regularization to improve self-supervised representations, outperforming existing methods in robustness and out-of-distribution generalization.

Jovana Mitrovic, Brian McWilliams, Jacob Walker et al.

Proposes non-leaky single-spike temporal coding SNN, analyzes input-output relations, applies to intrusion detection, outperforming DNNs with 98% accuracy.

Shibo Zhou, Xiaohua Li

Proposes Gradient Sign Dropout (GradDrop), a probabilistic masking method to resolve gradient conflicts, improving multitask and transfer learning performance.

Zhao Chen, Jiquan Ngiam, Yanping Huang et al.

Proposes Laplace-based LDP Bayesian optimization algorithms achieving near-optimal regret bounds.

Xingyu Zhou, Jian Tan

Proposes Informedness and Markedness as bias-neutral metrics, improving evaluation objectivity.

David M. W. Powers

Monte-Carlo simulations reveal 66% reproducibility for ICLR 2020, highlighting institutional bias and gender disparities.

David Tran, Alex Valtchanov, Keshav Ganapathy et al.

MeshGraphNets combines graph neural networks with adaptive remeshing for accurate, scalable physics simulation, outperforming traditional methods with 1-2 orders of magnitude speedup.

Tobias Pfaff, Meire Fortunato, Alvaro Sanchez-Gonzalez et al.

DreamerV2 uses discrete latent world models to achieve human-level Atari performance, surpassing top single-GPU algorithms with high sample efficiency.

Danijar Hafner, Timothy Lillicrap, Mohammad Norouzi et al.

AdaLead: a simple adaptive greedy algorithm for efficient biological sequence design, outperforming complex methods.

Sam Sinai, Richard Wang, Alexander Whatley et al.

Theoretically proves over-parameterization accelerates gradient descent convergence by shrinking the distance to the global optimum faster, based on a single teacher neuron model.

Jun-Kun Wang, Jacob Abernethy

Performer introduces a linear-time Transformer using FAVOR+ to unbiasedly approximate softmax attention, enabling scalable long-sequence modeling.

Krzysztof Choromanski, Valerii Likhosherstov, David Dohan et al.

This paper critically examines flaws in current time series anomaly detection benchmarks and introduces the UCR Anomaly Archive for more reliable evaluation.

Renjie Wu, Eamonn J. Keogh