Prescribing the Right Remedy: Mitigating Hallucinations in Large Vision-Language Models via Targeted Instruction Tuning

Despite achieving outstanding performance on various cross-modal tasks, current large vision-language models (LVLMs) still suffer from hallucination issues, manifesting as inconsistencies between their generated responses and the corresponding images. Prior research has implicated that the low quality of instruction data, particularly the skewed balance between positive and negative samples, is a significant contributor to model hallucinations. Recently, researchers have proposed high-quality instruction datasets, such as LRV-Instruction, to mitigate model hallucination. Nonetheless, our investigation reveals that hallucinatory concepts from different LVLMs exhibit specificity, i.e. the distribution of hallucinatory concepts varies significantly across models. Existing datasets did not consider the hallucination specificity of different models in the design processes, thereby diminishing their efficacy in mitigating model hallucination. In this paper, we propose a targeted instruction data generation framework named DFTG that tailored to the hallucination specificity of different models. Concretely, DFTG consists of two stages: hallucination diagnosis, which extracts the necessary information from the model's responses and images for hallucination diagnosis; and targeted data generation, which generates targeted instruction data based on diagnostic results. The experimental results on hallucination benchmarks demonstrate that the targeted instruction data generated by our method are more effective in mitigating hallucinations compared to previous datasets.

Implicit Image-to-Image Schrodinger Bridge for CT Super-Resolution and Denoising

Conditional diffusion models have gained recognition for their effectiveness in image restoration tasks, yet their iterative denoising process, starting from Gaussian noise, often leads to slow inference speeds. As a promising alternative, the Image-to-Image Schr\"odinger Bridge (I2SB) initializes the generative process from corrupted images and integrates training techniques from conditional diffusion models. In this study, we extended the I2SB method by introducing the Implicit Image-to-Image Schrodinger Bridge (I3SB), transitioning its generative process to a non-Markovian process by incorporating corrupted images in each generative step. This enhancement empowers I3SB to generate images with better texture restoration using a small number of generative steps. The proposed method was validated on CT super-resolution and denoising tasks and outperformed existing methods, including the conditional denoising diffusion probabilistic model (cDDPM) and I2SB, in both visual quality and quantitative metrics. These findings underscore the potential of I3SB in improving medical image restoration by providing fast and accurate generative modeling.

A Safe Reinforcement Learning Algorithm for Supervisory Control of Power Plants

Traditional control theory-based methods require tailored engineering for each system and constant fine-tuning. In power plant control, one often needs to obtain a precise representation of the system dynamics and carefully design the control scheme accordingly. Model-free Reinforcement learning (RL) has emerged as a promising solution for control tasks due to its ability to learn from trial-and-error interactions with the environment. It eliminates the need for explicitly modeling the environment's dynamics, which is potentially inaccurate. However, the direct imposition of state constraints in power plant control raises challenges for standard RL methods. To address this, we propose a chance-constrained RL algorithm based on Proximal Policy Optimization for supervisory control. Our method employs Lagrangian relaxation to convert the constrained optimization problem into an unconstrained objective, where trainable Lagrange multipliers enforce the state constraints. Our approach achieves the smallest distance of violation and violation rate in a load-follow maneuver for an advanced Nuclear Power Plant design.

Language-assisted Vision Model Debugger: A Sample-Free Approach to Finding Bugs

Vision models with high overall accuracy often exhibit systematic errors in specific scenarios, posing potential serious safety concerns. Diagnosing bugs of vision models is gaining increased attention, however traditional diagnostic approaches require annotation efforts (\eg rich metadata accompanying each samples of CelebA). To address this issue,We propose a language-assisted diagnostic method that uses texts instead of images to diagnose bugs in vision models based on multi-modal models (\eg CLIP). Our approach connects the embedding space of CLIP with the buggy vision model to be diagnosed; meanwhile, utilizing a shared classifier and the cross-modal transferability of embedding space from CLIP, the text-branch of CLIP become a proxy model to find bugs in the buggy model. The proxy model can classify texts paired with images. During the diagnosis, a Large Language Model (LLM) is employed to obtain task-relevant corpora, and this corpora is used to extract keywords. Descriptions constructed with templates containing these keywords serve as input text to probe errors in the proxy model. Finally, we validate the ability to diagnose existing visual models using language on the Waterbirds and CelebA datasets, we can identify bugs comprehensible to human experts, uncovering not only known bugs but also previously unknown ones.

Reconstructing Objects in-the-wild for Realistic Sensor Simulation

Reconstructing objects from real world data and rendering them at novel views is critical to bringing realism, diversity and scale to simulation for robotics training and testing. In this work, we present NeuSim, a novel approach that estimates accurate geometry and realistic appearance from sparse in-the-wild data captured at distance and at limited viewpoints. Towards this goal, we represent the object surface as a neural signed distance function and leverage both LiDAR and camera sensor data to reconstruct smooth and accurate geometry and normals. We model the object appearance with a robust physics-inspired reflectance representation effective for in-the-wild data. Our experiments show that NeuSim has strong view synthesis performance on challenging scenarios with sparse training views. Furthermore, we showcase composing NeuSim assets into a virtual world and generating realistic multi-sensor data for evaluating self-driving perception models.

Learning Unsupervised World Models for Autonomous Driving via Discrete Diffusion

Learning world models can teach an agent how the world works in an unsupervised manner. Even though it can be viewed as a special case of sequence modeling, progress for scaling world models on robotic applications such as autonomous driving has been somewhat less rapid than scaling language models with Generative Pre-trained Transformers (GPT). We identify two reasons as major bottlenecks: dealing with complex and unstructured observation space, and having a scalable generative model. Consequently, we propose a novel world modeling approach that first tokenizes sensor observations with VQVAE, then predicts the future via discrete diffusion. To efficiently decode and denoise tokens in parallel, we recast Masked Generative Image Transformer into the discrete diffusion framework with a few simple changes, resulting in notable improvement. When applied to learning world models on point cloud observations, our model reduces prior SOTA Chamfer distance by more than 65% for 1s prediction, and more than 50% for 3s prediction, across NuScenes, KITTI Odometry, and Argoverse2 datasets. Our results demonstrate that discrete diffusion on tokenized agent experience can unlock the power of GPT-like unsupervised learning for robotic agents.

Skywork: A More Open Bilingual Foundation Model

In this technical report, we present Skywork-13B, a family of large language models (LLMs) trained on a corpus of over 3.2 trillion tokens drawn from both English and Chinese texts. This bilingual foundation model is the most extensively trained and openly published LLMs of comparable size to date. We introduce a two-stage training methodology using a segmented corpus, targeting general purpose training and then domain-specific enhancement training, respectively. We show that our model not only excels on popular benchmarks, but also achieves \emph{state of the art} performance in Chinese language modeling on diverse domains. Furthermore, we propose a novel leakage detection method, demonstrating that test data contamination is a pressing issue warranting further investigation by the LLM community. To spur future research, we release Skywork-13B along with checkpoints obtained during intermediate stages of the training process. We are also releasing part of our SkyPile corpus, a collection of over 150 billion tokens of web text, which is the largest high quality open Chinese pre-training corpus to date. We hope Skywork-13B and our open corpus will serve as a valuable open-source resource to democratize access to high-quality LLMs.

SkyMath: Technical Report

Large language models (LLMs) have shown great potential to solve varieties of natural language processing (NLP) tasks, including mathematical reasoning. In this work, we present SkyMath, a large language model for mathematics with 13 billion parameters. By applying self-compare fine-tuning, we have enhanced mathematical reasoning abilities of Skywork-13B-Base remarkably. On GSM8K, SkyMath outperforms all known open-source models of similar size and has established a new SOTA performance.

Byzantine-Robust Federated Learning with Variance Reduction and Differential Privacy

Federated learning (FL) is designed to preserve data privacy during model training, where the data remains on the client side (i.e., IoT devices), and only model updates of clients are shared iteratively for collaborative learning. However, this process is vulnerable to privacy attacks and Byzantine attacks: the local model updates shared throughout the FL network will leak private information about the local training data, and they can also be maliciously crafted by Byzantine attackers to disturb the learning. In this paper, we propose a new FL scheme that guarantees rigorous privacy and simultaneously enhances system robustness against Byzantine attacks. Our approach introduces sparsification- and momentum-driven variance reduction into the client-level differential privacy (DP) mechanism, to defend against Byzantine attackers. The security design does not violate the privacy guarantee of the client-level DP mechanism; hence, our approach achieves the same client-level DP guarantee as the state-of-the-art. We conduct extensive experiments on both IID and non-IID datasets and different tasks and evaluate the performance of our approach against different Byzantine attacks by comparing it with state-of-the-art defense methods. The results of our experiments show the efficacy of our framework and demonstrate its ability to improve system robustness against Byzantine attacks while achieving a strong privacy guarantee.

Rethinking Closed-loop Training for Autonomous Driving

Recent advances in high-fidelity simulators have enabled closed-loop training of autonomous driving agents, potentially solving the distribution shift in training v.s. deployment and allowing training to be scaled both safely and cheaply. However, there is a lack of understanding of how to build effective training benchmarks for closed-loop training. In this work, we present the first empirical study which analyzes the effects of different training benchmark designs on the success of learning agents, such as how to design traffic scenarios and scale training environments. Furthermore, we show that many popular RL algorithms cannot achieve satisfactory performance in the context of autonomous driving, as they lack long-term planning and take an extremely long time to train. To address these issues, we propose trajectory value learning (TRAVL), an RL-based driving agent that performs planning with multistep look-ahead and exploits cheaply generated imagined data for efficient learning. Our experiments show that TRAVL can learn much faster and produce safer maneuvers compared to all the baselines. For more information, visit the project website: https://waabi.ai/research/travl