This paper surveys research in the rapidly growing field of enhancing large language models (LLMs) with reinforcement learning (RL), a technique that enables LLMs to improve their performance by receiving feedback in the form of rewards based on the quality of their outputs, allowing them to generate more accurate, coherent, and contextually appropriate responses. In this work, we make a systematic review of the most up-to-date state of knowledge on RL-enhanced LLMs, attempting to consolidate and analyze the rapidly growing research in this field, helping researchers understand the current challenges and advancements. Specifically, we (1) detail the basics of RL; (2) introduce popular RL-enhanced LLMs; (3) review researches on two widely-used reward model-based RL techniques: Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning from AI Feedback (RLAIF); and (4) explore Direct Preference Optimization (DPO), a set of methods that bypass the reward model to directly use human preference data for aligning LLM outputs with human expectations. We will also point out current challenges and deficiencies of existing methods and suggest some avenues for further improvements. Project page of this work can be found at: \url{https://github.com/ShuheWang1998/Reinforcement-Learning-Enhanced-LLMs-A-Survey}.

Few-shot Class-Incremental Learning (FSCIL) addresses the challenges of evolving data distributions and the difficulty of data acquisition in real-world scenarios. To counteract the catastrophic forgetting typically encountered in FSCIL, knowledge distillation is employed as a way to maintain the knowledge from learned data distribution. Recognizing the limitations of generating discriminative feature representations in a few-shot context, our approach incorporates structural information between samples into knowledge distillation. This structural information serves as a remedy for the low quality of features. Diverging from traditional structured distillation methods that compute sample similarity, we introduce the Displacement Knowledge Distillation (DKD) method. DKD utilizes displacement rather than similarity between samples, incorporating both distance and angular information to significantly enhance the information density retained through knowledge distillation. Observing performance disparities in feature distribution between base and novel classes, we propose the Dual Distillation Network (DDNet). This network applies traditional knowledge distillation to base classes and DKD to novel classes, challenging the conventional integration of novel classes with base classes. Additionally, we implement an instance-aware sample selector during inference to dynamically adjust dual branch weights, thereby leveraging the complementary strengths of each approach. Extensive testing on three benchmarks demonstrates that DDNet achieves state-of-the-art results. Moreover, through rigorous experimentation and comparison, we establish the robustness and general applicability of our proposed DKD method.

Quantum Boltzmann machines (QBMs) are machine-learning models for both classical and quantum data. We give an operational definition of QBM learning in terms of the difference in expectation values between the model and target, taking into account the polynomial size of the data set. By using the relative entropy as a loss function this problem can be solved without encountering barren plateaus. We prove that a solution can be obtained with stochastic gradient descent using at most a polynomial number of Gibbs states. We also prove that pre-training on a subset of the QBM parameters can only lower the sample complexity bounds. In particular, we give pre-training strategies based on mean-field, Gaussian Fermionic, and geometrically local Hamiltonians. We verify these models and our theoretical findings numerically on a quantum and a classical data set. Our results establish that QBMs are promising machine learning models.

Knowledge graphs are an efficient method for representing and connecting information across various concepts, useful in reasoning, question answering, and knowledge base completion tasks. They organize data by linking points, enabling researchers to combine diverse information sources into a single database. This interdisciplinary approach helps uncover new research questions and ideas. Knowledge graphs create a web of data points (nodes) and their connections (edges), which enhances navigation, comprehension, and utilization of data for multiple purposes. They capture complex relationships inherent in unstructured data sources, offering a semantic framework for diverse entities and their attributes. Strategies for developing knowledge graphs include using seed data, named entity recognition, and relationship extraction. These graphs enhance chatbot accuracy and include multimedia data for richer information. Creating high-quality knowledge graphs involves both automated methods and human oversight, essential for accurate and comprehensive data representation.

The association of social mobility with a college education has been studied since the early 1950's [1]. Although there are some indications that a college education is not as effective as it once was in helping graduates climb the social ladder [2], it is still the most reliable way of doing so. US News & World Report updated its rankings in 2023 to include social mobility [3], and many institutions of higher education are paying more attention to recruitment of first-generation college students and talented students from disadvantaged backgrounds. With the inclusion of such students in the typical college class comes some important considerations. For example, a student from difficult financial circumstances with an academic background to match the profile of any student an elite institution will have more difficulty paying for textbooks, a laptop, a smartphone, and other items that are almost essential to current college life [2]. As of November 2022, one such item that students from advantaged backgrounds will have access to that those from lower income brackets will not is ChatGPT4 [4]. It currently costs $20 per month for a subscription and has been called a "significant leap forward" compared to ChatGPT3.5 [5], which is free [6]. While use of generative AI is prohibited in some college classrooms, this is hard to police, and many students use it regardless of classroom restrictions [7]. When generative AI is allowed, there is a wide array of platforms from which students can choose.

Low-Rank Adaptation (LoRA) is a parameter-efficient fine-tuning method that has been widely adopted in various downstream applications of LLMs. Together with the Mixture-of-Expert (MoE) technique, fine-tuning approaches have shown remarkable improvements in model capability. However, the coordination of multiple experts in existing studies solely relies on the weights assigned by the simple router function. Lack of communication and collaboration among experts exacerbate the instability of LLMs due to the imbalance load problem of MoE. To address this issue, we propose a novel MoE graph-based LLM fine-tuning framework GraphLoRA, in which a graph router function is designed to capture the collaboration signals among experts by graph neural networks (GNNs). GraphLoRA enables all experts to understand input knowledge and share information from neighbor experts by aggregating operations. Besides, to enhance each expert's capability and their collaborations, we design two novel coordination strategies: the Poisson distribution-based distinction strategy and the Normal distribution-based load balance strategy. Extensive experiments on four real-world datasets demonstrate the effectiveness of our GraphLoRA in parameter-efficient fine-tuning of LLMs, showing the benefits of facilitating collaborations of multiple experts in the graph router of GraphLoRA.

Reliable responses from large language models (LLMs) require adherence to user instructions and retrieved information. While alignment techniques help LLMs align with human intentions and values, improving context-faithfulness through alignment remains underexplored. To address this, we propose $\textbf{Context-DPO}$, the first alignment method specifically designed to enhance LLMs' context-faithfulness. We introduce $\textbf{ConFiQA}$, a benchmark that simulates Retrieval-Augmented Generation (RAG) scenarios with knowledge conflicts to evaluate context-faithfulness. By leveraging faithful and stubborn responses to questions with provided context from ConFiQA, our Context-DPO aligns LLMs through direct preference optimization. Extensive experiments demonstrate that our Context-DPO significantly improves context-faithfulness, achieving 35% to 280% improvements on popular open-source models. Further analysis demonstrates that Context-DPO preserves LLMs' generative capabilities while providing interpretable insights into context utilization. Our code and data are released at https://github.com/byronBBL/Context-DPO

The deployment of artificial intelligence (AI) in critical decision-making and evaluation processes raises concerns about inherent biases that malicious actors could exploit to distort decision outcomes. We propose a systematic method to reveal such biases in AI evaluation systems and apply it to automated essay grading as an example. Our approach first identifies hidden neural activity patterns that predict distorted decision outcomes and then optimizes an adversarial input suffix to amplify such patterns. We demonstrate that this combination can effectively fool large language model (LLM) graders into assigning much higher grades than humans would. We further show that this white-box attack transfers to black-box attacks on other models, including commercial closed-source models like Gemini. They further reveal the existence of a "magic word" that plays a pivotal role in the efficacy of the attack. We trace the origin of this magic word bias to the structure of commonly-used chat templates for supervised fine-tuning of LLMs and show that a minor change in the template can drastically reduce the bias. This work not only uncovers vulnerabilities in current LLMs but also proposes a systematic method to identify and remove hidden biases, contributing to the goal of ensuring AI safety and security.

Human tutoring interventions play a crucial role in supporting student learning, improving academic performance, and promoting personal growth. This paper focuses on analyzing mathematics tutoring discourse using talk moves - a framework of dialogue acts grounded in Accountable Talk theory. However, scaling the collection, annotation, and analysis of extensive tutoring dialogues to develop machine learning models is a challenging and resource-intensive task. To address this, we present SAGA22, a compact dataset, and explore various modeling strategies, including dialogue context, speaker information, pretraining datasets, and further fine-tuning. By leveraging existing datasets and models designed for classroom teaching, our results demonstrate that supplementary pretraining on classroom data enhances model performance in tutoring settings, particularly when incorporating longer context and speaker information. Additionally, we conduct extensive ablation studies to underscore the challenges in talk move modeling.

Online learning of deep neural networks suffers from challenges such as hysteretic non-incremental updating, increasing memory usage, past retrospective retraining, and catastrophic forgetting. To alleviate these drawbacks and achieve progressive immediate decision-making, we propose a novel Incremental Online Learning (IOL) process of Randomized Neural Networks (Randomized NN), a framework facilitating continuous improvements to Randomized NN performance in restrictive online scenarios. Within the framework, we further introduce IOL with ridge regularization (-R) and IOL with forward regularization (-F). -R generates stepwise incremental updates without retrospective retraining and avoids catastrophic forgetting. Moreover, we substituted -R with -F as it enhanced precognition learning ability using semi-supervision and realized better online regrets to offline global experts compared to -R during IOL. The algorithms of IOL for Randomized NN with -R/-F on non-stationary batch stream were derived respectively, featuring recursive weight updates and variable learning rates. Additionally, we conducted a detailed analysis and theoretically derived relative cumulative regret bounds of the Randomized NN learners with -R/-F in IOL under adversarial assumptions using a novel methodology and presented several corollaries, from which we observed the superiority on online learning acceleration and regret bounds of employing -F in IOL. Finally, our proposed methods were rigorously examined across regression and classification tasks on diverse datasets, which distinctly validated the efficacy of IOL frameworks of Randomized NN and the advantages of forward regularization.