Search

Reinforcement learning from human feedback (RLHF) is a prevalent approach to align AI systems with human values by learning rewards from human preference data. Due to various reasons, however, such data typically takes the form of rankings over pairs of trajectory segments, which fails to capture the varying strengths of preferences across different pairs. In this paper, we propose a novel adaptive preference loss, underpinned by distributionally robust optimization (DRO), designed to address this uncertainty in preference strength. By incorporating an adaptive scaling parameter into the loss for each pair, our method increases the flexibility of the reward function. Specifically, it assigns small scaling parameters to pairs with ambiguous preferences, leading to more comparable rewards, and large scaling parameters to those with clear preferences for more distinct rewards. Computationally, our proposed loss function is strictly convex and univariate with respect to each scaling parameter, enabling its efficient optimization through a simple second-order algorithm. Our method is versatile and can be readily adapted to various preference optimization frameworks, including direct preference optimization (DPO). Our experiments with robotic control and natural language generation with large language models (LLMs) show that our method not only improves policy performance but also aligns reward function selection more closely with policy optimization, simplifying the hyperparameter tuning process.

Efficiency and reliability are critical in robotic bin-picking as they directly impact the productivity of automated industrial processes. However, traditional approaches, demanding static objects and fixed collisions, lead to deployment limitations, operational inefficiencies, and process unreliability. This paper introduces a Dynamic Bin-Picking Framework (DBPF) that challenges traditional static assumptions. The DBPF endows the robot with the reactivity to pick multiple moving arbitrary objects while avoiding dynamic obstacles, such as the moving bin. Combined with scene-level pose generation, the proposed pose selection metric leverages the Tendency-Aware Manipulability Network optimizing suction pose determination. Heuristic task-specific designs like velocity-matching, dynamic obstacle avoidance, and the resight policy, enhance the picking success rate and reliability. Empirical experiments demonstrate the importance of these components. Our method achieves an average 84% success rate, surpassing the 60% of the most comparable baseline, crucially, with zero collisions. Further evaluations under diverse dynamic scenarios showcase DBPF's robust performance in dynamic bin-picking. Results suggest that our framework offers a promising solution for efficient and reliable robotic bin-picking under dynamics., Comment: 8 pages, 5 figures. This paper has been accepted by IEEE RA-L on 2024-03-24. See the supplementary video at youtube: https://youtu.be/n5af2VsKhkg

There has emerged a growing interest in exploring efficient quality assessment algorithms for image super-resolution (SR). However, employing deep learning techniques, especially dual-branch algorithms, to automatically evaluate the visual quality of SR images remains challenging. Existing SR image quality assessment (IQA) metrics based on two-stream networks lack interactions between branches. To address this, we propose a novel full-reference IQA (FR-IQA) method for SR images. Specifically, producing SR images and evaluating how close the SR images are to the corresponding HR references are separate processes. Based on this consideration, we construct a deep Bi-directional Attention Network (BiAtten-Net) that dynamically deepens visual attention to distortions in both processes, which aligns well with the human visual system (HVS). Experiments on public SR quality databases demonstrate the superiority of our proposed BiAtten-Net over state-of-the-art quality assessment methods. In addition, the visualization results and ablation study show the effectiveness of bi-directional attention., Comment: 7 pages, 3 figures, published to 2024 IEEE International Conference on Multimedia and Expo (ICME)

Under the global low-carbon target, the uneven spatiotemporal distribution of renewable energy resources exacerbates the uncertainty and seasonal power imbalance. Additionally, the issue of an incomplete hydrogen energy chain is widely overlooked in planning models, which hinders the complete analysis of the role of hydrogen in energy systems. Therefore, this paper proposes a high-resolution collaborative planning model for electricity-thermal-hydrogen-coupled energy systems considering both the spatiotemporal distribution characteristics of renewable energy resources and the multi-scale bottom-to-top investment strategy for the complete hydrogen energy chain. Considering the high-resolution system operation flexibility, this paper proposes a hydrogen chain-based fast clustering optimization method that can handle high-dimensional data and multi-time scale operation characteristics. The model optimizes the geographical distribution and capacity configuration of the Northeast China energy system in 2050, with hourly operational characteristics. The planning optimization covered single-energy devices, multi-energy-coupled conversion devices, and electric-hydrogen transmission networks. Last but not least, this paper thoroughly examines the optimal portfolio selection of different hydrogen technologies based on the differences in cost, flexibility, and efficiency. In the Pareto analysis, the proposed model reduces CO2 emissions by 60% with a competitive cost. This paper provides a zero-carbon pathway for multi-energy systems with a cost 4% less than the social cost of carbon $44.6/ton, and the integration of the complete hydrogen energy chain reduces the renewable energy curtailment by 97.0%. Besides, the portfolio selection results indicate that the system favors the SOEC with the highest energy efficiency and the PEMFC with the fastest dynamic response when achieving zero-carbon emissions, Comment: 32 pages, 17 figures

Quantization is an indispensable technique for serving Large Language Models (LLMs) and has recently found its way into LoRA fine-tuning. In this work we focus on the scenario where quantization and LoRA fine-tuning are applied together on a pre-trained model. In such cases it is common to observe a consistent gap in the performance on downstream tasks between full fine-tuning and quantization plus LoRA fine-tuning approach. In response, we propose LoftQ (LoRA-Fine-Tuning-aware Quantization), a novel quantization framework that simultaneously quantizes an LLM and finds a proper low-rank initialization for LoRA fine-tuning. Such an initialization alleviates the discrepancy between the quantized and full-precision model and significantly improves the generalization in downstream tasks. We evaluate our method on natural language understanding, question answering, summarization, and natural language generation tasks. Experiments show that our method is highly effective and outperforms existing quantization methods, especially in the challenging 2-bit and 2/4-bit mixed precision regimes. We will release our code.

Reward design is a fundamental, yet challenging aspect of reinforcement learning (RL). Researchers typically utilize feedback signals from the environment to handcraft a reward function, but this process is not always effective due to the varying scale and intricate dependencies of the feedback signals. This paper shows by exploiting certain structures, one can ease the reward design process. Specifically, we propose a hierarchical reward modeling framework -- HERON for scenarios: (I) The feedback signals naturally present hierarchy; (II) The reward is sparse, but with less important surrogate feedback to help policy learning. Both scenarios allow us to design a hierarchical decision tree induced by the importance ranking of the feedback signals to compare RL trajectories. With such preference data, we can then train a reward model for policy learning. We apply HERON to several RL applications, and we find that our framework can not only train high performing agents on a variety of difficult tasks, but also provide additional benefits such as improved sample efficiency and robustness. Our code is available at \url{https://github.com/abukharin3/HERON}., Comment: 28 Pages, 14 figures

Transformer models have achieved remarkable results in various natural language tasks, but they are often prohibitively large, requiring massive memories and computational resources. To reduce the size and complexity of these models, we propose LoSparse (Low-Rank and Sparse approximation), a novel model compression technique that approximates a weight matrix by the sum of a low-rank matrix and a sparse matrix. Our method combines the advantages of both low-rank approximations and pruning, while avoiding their limitations. Low-rank approximation compresses the coherent and expressive parts in neurons, while pruning removes the incoherent and non-expressive parts in neurons. Pruning enhances the diversity of low-rank approximations, and low-rank approximation prevents pruning from losing too many expressive neurons. We evaluate our method on natural language understanding, question answering, and natural language generation tasks. We show that it significantly outperforms existing compression methods.

Freezing interface interacting with soft cells is a core issue in cryopreservation. However, most of existing cryopreservation approaches face challenges such as complex processing and poor controllability. Herein we report a new, nanotechnology-based method to modulate interfacial interactions between ice and cells via the premelting theory. Through the interfacial design by controllably modifying biocompatible nanoshells on the surface of cells, the effective Hamaker constant between cells and ice can be modified. The thickness of premelted films between cells and ice is further regulated, and directional migration of coated cells occurs in a thermal gradient to achieve directional thawing of coated cells. The decreased mortality of freezing coated cells suggests the premelting modulation in freezing and thawing can effectively protect cells from mechanical damage of ice formation during freezing and ice recrystallization during thawing.

The objective of the change-point detection is to discover the abrupt property changes lying behind the time-series data. In this paper, we firstly summarize the definition and in-depth implication of the changepoint detection. The next stage is to elaborate traditional and some alternative model-based changepoint detection algorithms. Finally, we try to go a bit further in the theory and look into future research directions., Comment: 11 pages