The success of AI assistants based on Language Models (LLMs) hinges on Reinforcement Learning from Human Feedback (RLHF) to comprehend and align with user intentions. However, traditional alignment algorithms, such as PPO, are hampered by complex annotation and training requirements. This reliance limits the applicability of RLHF and hinders the development of professional assistants tailored to diverse human preferences. In this work, we introduce \textit{Linear Alignment}, a novel algorithm that aligns language models with human preferences in one single inference step, eliminating the reliance on data annotation and model training. Linear alignment incorporates a new parameterization for policy optimization under divergence constraints, which enables the extraction of optimal policy in a closed-form manner and facilitates the direct estimation of the aligned response. Extensive experiments on both general and personalized preference datasets demonstrate that linear alignment significantly enhances the performance and efficiency of LLM alignment across diverse scenarios. Our code and dataset will be published on \url{https://github.com/Wizardcoast/Linear_Alignment.git}.
This paper studies the computational offloading of CNN inference in dynamic multi-access edge computing (MEC) networks. To address the uncertainties in communication time and computation resource availability, we propose a novel semantic compression method, autoencoder-based CNN architecture (AECNN), for effective semantic extraction and compression in partial offloading. In the semantic encoder, we introduce a feature compression module based on the channel attention mechanism in CNNs, to compress intermediate data by selecting the most informative features. In the semantic decoder, we design a lightweight decoder to reconstruct the intermediate data through learning from the received compressed data to improve accuracy. To effectively trade-off communication, computation, and inference accuracy, we design a reward function and formulate the offloading problem of CNN inference as a maximization problem with the goal of maximizing the average inference accuracy and throughput over the long term. To address this maximization problem, we propose a graph reinforcement learning-based AECNN (GRL-AECNN) method, which outperforms existing works DROO-AECNN, GRL-BottleNet++ and GRL-DeepJSCC under different dynamic scenarios. This highlights the advantages of GRL-AECNN in offloading decision-making in dynamic MEC.
Tool learning has generated widespread interest as a vital means of interaction between Large Language Models (LLMs) and the physical world. Current research predominantly emphasizes LLMs' capacity to utilize tools in well-structured environments while overlooking their stability when confronted with the inevitable noise of the real world. To bridge this gap, we introduce RoTBench, a multi-level benchmark for evaluating the robustness of LLMs in tool learning. Specifically, we establish five external environments, each featuring varying levels of noise (i.e., Clean, Slight, Medium, Heavy, and Union), providing an in-depth analysis of the model's resilience across three critical phases: tool selection, parameter identification, and content filling. Experiments involving six widely-used models underscore the urgent necessity for enhancing the robustness of LLMs in tool learning. For instance, the performance of GPT-4 even drops significantly from 80.00 to 58.10 when there is no substantial change in manual accuracy. More surprisingly, the noise correction capability inherent in the GPT family paradoxically impedes its adaptability in the face of mild noise. In light of these findings, we propose RoTTuning, a strategy that enriches the diversity of training environments to bolster the robustness of LLMs in tool learning. The code and data are available at https://github.com/Junjie-Ye/RoTBench.
To enhance the domain-specific capabilities of large language models, continued pre-training on a domain-specific corpus is a prevalent method. Recent work demonstrates that adapting models using reading comprehension data formatted by regex-based patterns can significantly improve performance on domain-specific tasks. However, regex-based patterns are incapable of parsing raw corpora using domain-specific knowledge. Furthermore, the question and answer pairs are extracted directly from the corpus in predefined formats offers limited context. To address this limitation, we improve reading comprehension via LLM and clustering. LLM focuses on leveraging domain knowledge within the corpus to refine comprehension stage, while clustering supplies relevant knowledge by extending the context to enrich reading stage. Additionally, our method incorporates parameter-efficient fine-tuning to improve the efficiency of domain adaptation. In comparison to AdaptLLM, our method achieves an improvement exceeding 5% in domain-specific tasks. Our code will available at https://github.com/microsoft/LMOps.
Existing evaluations of tool learning primarily focus on validating the alignment of selected tools for large language models (LLMs) with expected outcomes. However, these approaches rely on a limited set of scenarios where answers can be pre-determined, diverging from genuine needs. Furthermore, a sole emphasis on outcomes disregards the intricate capabilities essential for LLMs to effectively utilize tools. To tackle this issue, we propose ToolEyes, a fine-grained system tailored for the evaluation of the LLMs' tool learning capabilities in authentic scenarios. The system meticulously examines seven real-world scenarios, analyzing five dimensions crucial to LLMs in tool learning: format alignment, intent comprehension, behavior planning, tool selection, and answer organization. Additionally, ToolEyes incorporates a tool library boasting approximately 600 tools, serving as an intermediary between LLMs and the physical world. Evaluations involving ten LLMs across three categories reveal a preference for specific scenarios and limited cognitive abilities in tool learning. Intriguingly, expanding the model size even exacerbates the hindrance to tool learning. These findings offer instructive insights aimed at advancing the field of tool learning. The data is available att https://github.com/Junjie-Ye/ToolEyes.
We address the challenge of learning safe and robust decision policies in presence of uncertainty in context of the real scientific problem of adaptive resource oversubscription to enhance resource efficiency while ensuring safety against resource congestion risk. Traditional supervised prediction or forecasting models are ineffective in learning adaptive policies whereas standard online optimization or reinforcement learning is difficult to deploy on real systems. Offline methods such as imitation learning (IL) are ideal since we can directly leverage historical resource usage telemetry. But, the underlying aleatoric uncertainty in such telemetry is a critical bottleneck. We solve this with our proposed novel chance-constrained imitation learning framework, which ensures implicit safety against uncertainty in a principled manner via a combination of stochastic (chance) constraints on resource congestion risk and ensemble value functions. This leads to substantial ($\approx 3-4\times$) improvement in resource efficiency and safety in many oversubscription scenarios, including resource management in cloud services.
As one of the most effective self-supervised representation learning methods, contrastive learning (CL) relies on multiple negative pairs to contrast against each positive pair. In the standard practice of contrastive learning, data augmentation methods are utilized to generate both positive and negative pairs. While existing works have been focusing on improving the positive sampling, the negative sampling process is often overlooked. In fact, the generated negative samples are often polluted by positive samples, which leads to a biased loss and performance degradation. To correct the negative sampling bias, we propose a novel contrastive learning method named Positive-Unlabeled Contrastive Learning (PUCL). PUCL treats the generated negative samples as unlabeled samples and uses information from positive samples to correct bias in contrastive loss. We prove that the corrected loss used in PUCL only incurs a negligible bias compared to the unbiased contrastive loss. PUCL can be applied to general contrastive learning problems and outperforms state-of-the-art methods on various image and graph classification tasks. The code of PUCL is in the supplementary file.
Reinforcement Learning from Human Feedback (RLHF) has become a crucial technology for aligning language models with human values and intentions, enabling models to produce more helpful and harmless responses. Reward models are trained as proxies for human preferences to drive reinforcement learning optimization. While reward models are often considered central to achieving high performance, they face the following challenges in practical applications: (1) Incorrect and ambiguous preference pairs in the dataset may hinder the reward model from accurately capturing human intent. (2) Reward models trained on data from a specific distribution often struggle to generalize to examples outside that distribution and are not suitable for iterative RLHF training. In this report, we attempt to address these two issues. (1) From a data perspective, we propose a method to measure the strength of preferences within the data, based on a voting mechanism of multiple reward models. Experimental results confirm that data with varying preference strengths have different impacts on reward model performance. We introduce a series of novel methods to mitigate the influence of incorrect and ambiguous preferences in the dataset and fully leverage high-quality preference data. (2) From an algorithmic standpoint, we introduce contrastive learning to enhance the ability of reward models to distinguish between chosen and rejected responses, thereby improving model generalization. Furthermore, we employ meta-learning to enable the reward model to maintain the ability to differentiate subtle differences in out-of-distribution samples, and this approach can be utilized for iterative RLHF optimization.