This paper introduces UniOcc, a new benchmark for 3D occupancy forecasting and prediction in autonomous driving. It unifies data from real-world datasets like nuScenes and Waymo with synthetic data from CARLA and OpenCOOD, providing 2D/3D occupancy labels and voxel-level flow. UniOcc also offers novel evaluation metrics that don't rely solely on imperfect ground truth and includes a toolkit for object segmentation and tracking within the voxel space, aiming to standardize and advance research in this critical area.

This research introduces a novel method for generating counterfactual explanations for graph-based predictions.The approach utilizes a permutation equivariant graph variational autoencoder (PEGVAE) to create a meaningful latent space representation of graphs. By traversing this latent space guided by the classifier, the method identifies minimally altered graphs with a different classification. This technique addresses the challenges posed by the discrete nature and permutation invariance of graphs. Experiments on molecular datasets demonstrate the method's ability to produce robust and high-performing counterfactual explanations compared to baseline approaches. The work offers a way to understand how to change a graph to alter its predicted outcome without explicitly defining a graph distance metric.

The provided research paper explores the novel problem of continual forgetting in pre-trained vision models, where the goal is to sequentially remove specific unwanted knowledge while retaining the model's performance on other tasks. To address this, the authors introduce Group Sparse LoRA (GS-LoRA++), a parameter-efficient fine-tuning approach that uses low-rank adaptation with a group sparsity regularization to selectively modify the model's feed-forward networks. Furthermore, to handle practical scenarios like scarce forgetting data, they incorporate prototype regularization to guide the forgetting process. Extensive experiments across various vision tasks demonstrate the effectiveness and efficiency of their proposed method in achieving selective and continual knowledge erasure with minimal impact on remaining knowledge.

The provided research explores optimizing communication and sensing for autonomous vehicles using millimeter wave technology and orthogonal frequency-division multiplexing. Facing dynamic environments and the impact of wireless conditions on mmWave, the authors propose an adaptive system. This system employs reinforcement learning, considering queue state and channel state information. The goal is to achieve reliable communication and accurate velocity estimation of surrounding objects. To improve performance, they introduce adaptive OFDM and a reward function that utilizes the age of updates. Simulation results using A2C and PPO algorithms demonstrate the superiority of their approach over existing designs.

Recent progress in video generation still struggles with issues like motion instability and prompt alignment. To address this, the study explores incorporating human preferences into advanced flow-based video generation models. The authors introduce a large, new dataset of human-annotated video preferences across visual quality, motion quality, and text alignment. They also develop a multi-dimensional reward model to quantify these preferences and propose three alignment algorithms for flow-based models, demonstrating that a modified Direct Preference Optimization method yields the most effective results in aligning video generation with human expectations.

This research paper introduces AnimeGamer, a novel method for creating infinite anime life simulation games. Unlike previous work limited to static images or finite gameplay, AnimeGamer uses a multimodal large language model to process user instructions and predict subsequent game states. These states include dynamic animation shots and updated character attributes, aiming for a more immersive and consistent gaming experience. The paper details the model's architecture, training process, and evaluation, demonstrating its superior performance in generating coherent and interactive anime game worlds compared to baseline methods.

The provided research introduces NoProp, a novel method for training neural networks that deviates from traditional backpropagation by eliminating both forward and backward passes. Instead, it draws inspiration from diffusion models, where each layer learns independently to denoise a noisy version of the target. This approach trains networks without hierarchical representation learning in the conventional sense, achieving competitive or superior accuracy and efficiency compared to existing backpropagation-free techniques on image classification tasks. The work explores discrete-time and continuous-time variations of NoProp, including connections to flow matching, and examines its performance and memory usage relative to backpropagation and other alternative optimization methods.

The provided paper introduces ACPBench Hard, a new benchmark designed to evaluate the reasoning capabilities of large language models for automated planning. This benchmark extends the original ACPBench by featuring open-ended, generative versions of planning-related questions across various tasks, mirroring the challenges faced by symbolic planners. The authors tested several large language and reasoning models, including state-of-the-art options, on ACPBench Hard and found that their performance was generally subpar, indicating a significant gap in their ability to reliably reason about planning. The research highlights the difficulty current models face with core planning reasoning tasks and suggests future directions for improving their performance.

From Amazon's blog

Dynamic 4D Modeling from Casual Video