Large-scale alignment pipelines typically pair a policy model with a separately trained reward model whose parameters remain frozen during reinforcement learning (RL). This separation creates a complex, resource-intensive pipeline and leads to a performance ceiling. We propose a novel framework, Unified Reward & Policy Optimization (URPO), that unifies instruction-following (“player”) and reward modeling (“referee”) into a single model and a single training phase. Our method recasts all alignment data-including preference pairs, verifiable reasoning, and open-ended instructions-into a unified generative format optimized by a single Group-Relative Policy Optimization (GRPO) loop. This enables the model to learn from ground-truth preferences and verifiable logic while simultaneously generating its own rewards for open-ended tasks. Experiments on the Qwen2.5-7B model demonstrate that URPO significantly outperforms a strong baseline using a separate generative reward model, boosting the instructionfollowing score on AlpacaEval to 44.84 and achieving a 36% relative improvement on the challenging AIME reasoning benchmark. Furthermore, URPO cultivates a superior internal evaluator as a byproduct of training, achieving a RewardBench score of 85.15 and surpassing the dedicated reward model it replaces (83.55). By eliminating the need for a separate reward model and fostering a co-evolutionary dynamic, URPO presents a simpler, more efficient, and more effective path towards robustly aligned language models.

Current Retrieval-Augmented Generation (RAG) systems concatenate and process numerous retrieved document chunks for prefill which requires a large volume of online computation, therefore leading to significant latency in time-to-first-token (TTFT).To reduce the computation overhead as well as TTFT, we introduce TurboRAG, a hybrid offline-online paradigm that (i) pre-computes chunk-level key-value (KV) caches, (ii) stitches them together at inference time using independent-attention and reordered-RoPE techniques, and (iii) preserves answer quality without changing the model architecture.Our approach is applicable to most existing large language models and their applications without any requirement in modification of models and inference systems.Experimental results across a suite of RAG benchmarks demonstrate that TurboRAG reduces TTFT by up to 9.4x compared to the conventional RAG systems (on an average of 8.6x), but reserving comparable performance to the standard RAG systems.

We present a generative dialogue system capable of operating in a full-duplex manner, allowing for seamless interaction. It is based on a large language model (LLM) carefully aligned to be aware of a perception module, a motor function module, and the concept of a simple finite state machine (called neural FSM) with two states. The perception and motor function modules operate in tandem, allowing the system to speak and listen to the user simultaneously. The LLM generates textual tokens for inquiry responses and makes autonomous decisions to start responding to, wait for, or interrupt the user by emitting control tokens to the neural FSM. All these tasks of the LLM are carried out as next token prediction on a serialized view of the dialogue in real-time. In automatic quality evaluations simulating real-life interaction, the proposed system reduces the average conversation response latency by more than threefold compared with LLM-based half-duplex dialogue systems while responding within less than 500 milliseconds in more than 50% of evaluated interactions. Running an LLM with only 8 billion parameters, our system exhibits an 8% higher interruption precision rate than the best available commercial LLM for voice-based dialogue.

In order to use remote wind power/solar energy, the scale of power grid interconnection grow rapidly. If transient stability can be increased when faults happen on tie lines, the transmission capacity can be increased too. Dynamic braking can increase transient stability by increasing deceleration area, but it's difficult to coordinate several braking devices and calculate braking amount in real time. Things are no longer so after rotor angle droop (RAD) controllers are deployed across the power system. Since all rotor angles of generators will be fixed in rotating coordination system determined by global position system (GPS) pulse per second (PPS) signal, dynamic braking amount can be calculated by RAD controller by using local speed and rotor measurement. This paper shows the revised equal area criteria (EAC) when this kind of dynamic braking is used. The relationships between RAD controller parameter (saturation threshold) and the braking effect are also investigated. Simulation results in IEEE 68 nodes system show that, by applying dynamic braking on several generators simultaneously, critical clearing time can be increased dramatically when faults happen on tie lines.

Support vector machine(SVM) has become a powerful and widely used machine learning method in resent years. Gaussian kernel is the most commonly used kernel function. However, model selection including setting the width parameter sigma in kernel function and the regularization parameter C is essential to generalization performance of SVM. In this paper we proposed a new parameter selection method for Support Vector Machine. The key idea of our method MSKD in selecting the Gaussian kernel parameter is that convergent character between pattern's similarity measurement in feature space will decrease the classification ability of SVM. In addition, We combined MSKD algorithm with one-dimension search strategy based on cross-validation and developed a complex parameters selection method named MSKD-GS. Experiments on eight real world data sets from UCI have been carried out to demonstrate the effectiveness and efficiency of this method.

Conventional ensemble learning algorithms based on ambiguity decomposition and negative correlation learning theory are carried out on the basis of empirical risk minimization principle. When SVM is used as the component learner, the generalization ability of ensemble learning system may not be improved. In this paper, based on the estimation of the generalization performance of SVM and negative correlation learning theory, a new selective ensemble SVM learning method is proposed. Experiments on real world data sets from UCI were carried out to demonstrate the effectiveness of this method.

Reduction of feature dimensionality is of considerable importance in machine learning. The generalization performance of classification system improves when correlated and redundant features are removed. In order to reduce the dimensionality of pattern representation, A new feature election method for support vector machine is proposed. Based on pattern similarity measurement in kernel space, lass separability is deduced and we explore the use of the lass separability in feature selection. The key idea of our ethod is that the feature whose removal downgrades the class separability in kernel space most is relevance to the classification. Experiments on linear and nonlinear synthetic problems and real (world data sets have been (carried out to demonstrate the effectiveness of this method.