Lost in Benchmarks? Rethinking Large Language Model Benchmarking with Item Response Theory

The evaluation of large language models (LLMs) via benchmarks is widespread, yet inconsistencies between different leaderboards and poor separability among top models raise concerns about their ability to accurately reflect authentic model capabilities. This paper provides a critical analysis of benchmark effectiveness, examining main-stream prominent LLM benchmarks using results from diverse models. We first propose a new framework for accurate and reliable estimations of item characteristics and model abilities. Specifically, we propose Pseudo-Siamese Network for Item Response Theory (PSN-IRT), an enhanced Item Response Theory framework that incorporates a rich set of item parameters within an IRT-grounded architecture. Based on PSN-IRT, we conduct extensive analysis which reveals significant and varied shortcomings in the measurement quality of current benchmarks. Furthermore, we demonstrate that leveraging PSN-IRT is able to construct smaller benchmarks while maintaining stronger alignment with human preference.

Generalized Neighborhood Attention: Multi-dimensional Sparse Attention at the Speed of Light

Many sparse attention mechanisms such as Neighborhood Attention have typically failed to consistently deliver speedup over the self attention baseline. This is largely due to the level of complexity in attention infrastructure, and the rapid evolution of AI hardware architecture. At the same time, many state-of-the-art foundational models, particularly in computer vision, are heavily bound by attention, and need reliable sparsity to escape the O(n^2) complexity. In this paper, we study a class of promising sparse attention mechanisms that focus on locality, and aim to develop a better analytical model of their performance improvements. We first introduce Generalized Neighborhood Attention (GNA), which can describe sliding window, strided sliding window, and blocked attention. We then consider possible design choices in implementing these approaches, and create a simulator that can provide much more realistic speedup upper bounds for any given setting. Finally, we implement GNA on top of a state-of-the-art fused multi-headed attention (FMHA) kernel designed for the NVIDIA Blackwell architecture in CUTLASS. Our implementation can fully realize the maximum speedup theoretically possible in many perfectly block-sparse cases, and achieves an effective utilization of 1.3 petaFLOPs/second in FP16. In addition, we plug various GNA configurations into off-the-shelf generative models, such as Cosmos-7B, HunyuanVideo, and FLUX, and show that it can deliver 28% to 46% end-to-end speedup on B200 without any fine-tuning. We will open source our simulator and Blackwell kernels directly through the NATTEN project.

SRLCG: Self-Rectified Large-Scale Code Generation with Multidimensional Chain-of-Thought and Dynamic Backtracking

Large language models (LLMs) have revolutionized code generation, significantly enhancing developer productivity. However, for a vast number of users with minimal coding knowledge, LLMs provide little support, as they primarily generate isolated code snippets rather than complete, large-scale project code. Without coding expertise, these users struggle to interpret, modify, and iteratively refine the outputs of LLMs, making it impossible to assemble a complete project. To address this issue, we propose Self-Rectified Large-Scale Code Generator (SRLCG), a framework that generates complete multi-file project code from a single prompt. SRLCG employs a novel multidimensional chain-of-thought (CoT) and self-rectification to guide LLMs in generating correct and robust code files, then integrates them into a complete and coherent project using our proposed dynamic backtracking algorithm. Experimental results show that SRLCG generates code 15x longer than DeepSeek-V3, 16x longer than GPT-4, and at least 10x longer than other leading CoT-based baselines. Furthermore, they confirm its improved correctness, robustness, and performance compared to baselines in large-scale code generation.

5G Direct Position Estimation for Precise Localization in Dense Urban Area

In recent years, the fifth-generation (5G) new radio (NR) signals have emerged as a promising supplementary resource for urban navigation. However, a major challenge in utilizing 5G signals lies in their vulnerability to non-line-of-sight (NLoS) propagation effects, which are especially prevalent in urban street canyons. This paper applies the direct position estimation (DPE) method to 5G cellular signals to mitigate the NLoS bias as well as the multipath effects, thereby enabling precise localization in urbanized environments. The feasibility of applying the DPE method to NR positioning is analyzed, followed by a discussion of the tapped delay line (TDL) channel propagation model provided by the 3rd Generation Partnership Project (3GPP). The positioning performance is then evaluated through large-scale system-level simulations. The simulation results demonstrate that 5G DPE achieves satisfactory positioning accuracy in a 10 dB noisy channel, with an overall root mean square error (RMSE) constrained within 6 m. In addition, 5G DPE outperforms the observed time difference of arrival (OTDoA) method by 95.24% in terms of positioning accuracy in an NLoS-dominated propagation environment.

MuSC: Improving Complex Instruction Following with Multi-granularity Self-Contrastive Training

Complex instruction-following with elaborate constraints is imperative for Large Language Models (LLMs). While existing methods have constructed data for complex instruction alignment, they all rely on a more advanced model, especially GPT-4, limiting their application. In this paper, we propose a Multi-granularity Self-Contrastive Training (MuSC) framework, to improve the complex instruction alignment without relying on a stronger model. Our method is conducted on both coarse and fine granularity. On coarse-granularity, we construct constraint-aware preference data based on instruction decomposition and recombination. On fine-granularity, we perform token-aware preference optimization with dynamic token-level supervision. Our method is evaluated on open-sourced models, and experiment results show our method achieves significant improvement on both complex and general instruction-following benchmarks, surpassing previous self-alignment methods.

Multipath Mitigation Technology-integrated GNSS Direct Position Estimation Plug-in Module

Direct position estimation (DPE) is an effective solution to the MP issue at the signal processing level. Unlike two-step positioning (2SP) receivers, DPE directly solves for the receiver position, velocity, and time (PVT) in the navigation domain, without the estimation of intermediate measurements, thus allowing it to provide more robust and accurate PVT estimates in the presence of multipath (MP) and weak signals. But GNSS positioning with DPE is mostly left unapplied commercially, and continuing research into DPE has remained relatively stagnant over the past few years. To encourage further research on DPE by the GNSS community, we propose a DPE plug-in module that can be integrated into the conventional 2SP software-defined receivers (SDRs). Programmed in MATLAB, the proposed DPE plug-in module is aimed for better understanding and familiarity of a practical implementation of DPE. Its plug-in module architecture allows it to be incorporated with 2SP MATLAB SDRs, both vector tracking and scalar tracking with minimum changes, making it easy to use, and provides greater flexibility for researchers using various 2SP SDRs. Since the proposed DPE implementation makes use of tracking observables from 2SP to propagate the channel, we propose to further improve the performance of DPE against MP through using MP-compensated observables generated from Multipath Mitigation Technology (MMT)-aided tracking. Referred to as Multipath Mitigation Technology (MMT)-integrated DPE, it is proposed as a variant of DPE that is better suit for urban environment applications. Results show that while in MP-only conditions, an MMT-integrated 2SP has similar performance with MMT-integrated DPE, the proposed MMT-integrated DPE manages to show great superiority against non-line-of-sight (NLOS), making it the preferable option for applications in urban environments.

PMoL: Parameter Efficient MoE for Preference Mixing of LLM Alignment

Reinforcement Learning from Human Feedback (RLHF) has been proven to be an effective method for preference alignment of large language models (LLMs) and is widely used in the post-training process of LLMs. However, RLHF struggles with handling multiple competing preferences. This leads to a decrease in the alignment of LLMs with human preferences. To address this issue, we propose Preference Mixture of LoRAs (PMoL) from the perspective of model architecture, which can adapt to any number of preferences to mix. PMoL combines Mixture of Experts (MoE) and Low Rank Adaptor (LoRA). This architecture is innovatively applied to the research of preference alignment and has achieved significant performance improvement. The expert group soft loss is used to enable MoE with the ability to mix preferences. Through comprehensive evaluation by the reward model and GPT-4o, the experiment results show that PMoL has superior preference mixing capabilities compared to baseline methods. PMoL achieves better preference alignment with lower training costs.

Mitigating the Bias of Large Language Model Evaluation

Recently, there has been a trend of evaluating the Large Language Model (LLM) quality in the flavor of LLM-as-a-Judge, namely leveraging another LLM to evaluate the current output quality. However, existing judges are proven to be biased, namely they would favor answers which present better superficial quality (such as verbosity, fluency) while ignoring the instruction following ability. In this work, we propose systematic research about the bias of LLM-as-a-Judge. Specifically, for closed-source judge models, we apply calibration to mitigate the significance of superficial quality, both on probability level and prompt level. For open-source judge models, we propose to mitigate the bias by contrastive training, with curated negative samples that deviate from instruction but present better superficial quality. We apply our methods on the bias evaluation benchmark, and experiment results show our methods mitigate the bias by a large margin while maintaining a satisfactory evaluation accuracy.

Navigating Public Sentiment in the Circular Economy through Topic Modelling and Hyperparameter Optimisation

To advance the circular economy (CE), it is crucial to gain insights into the evolution of public sentiments, cognitive pathways of the masses concerning circular products and digital technology, and recognise the primary concerns. To achieve this, we collected data related to the CE from diverse platforms including Twitter, Reddit, and The Guardian. This comprehensive data collection spanned across three distinct strata of the public: the general public, professionals, and official sources. Subsequently, we utilised three topic models on the collected data. Topic modelling represents a type of data-driven and machine learning approach for text mining, capable of automatically categorising a large number of documents into distinct semantic groups. Simultaneously, these groups are described by topics, and these topics can aid in understanding the semantic content of documents at a high level. However, the performance of topic modelling may vary depending on different hyperparameter values. Therefore, in this study, we proposed a framework for topic modelling with hyperparameter optimisation for CE and conducted a series of systematic experiments to ensure that topic models are set with appropriate hyperparameters and to gain insights into the correlations between the CE and public opinion based on well-established models. The results of this study indicate that concerns about sustainability and economic impact persist across all three datasets. Official sources demonstrate a higher level of engagement with the application and regulation of CE. To the best of our knowledge, this study is pioneering in investigating various levels of public opinions concerning CE through topic modelling with the exploration of hyperparameter optimisation.

Robust Causal Graph Representation Learning against Confounding Effects

The prevailing graph neural network models have achieved significant progress in graph representation learning. However, in this paper, we uncover an ever-overlooked phenomenon: the pre-trained graph representation learning model tested with full graphs underperforms the model tested with well-pruned graphs. This observation reveals that there exist confounders in graphs, which may interfere with the model learning semantic information, and current graph representation learning methods have not eliminated their influence. To tackle this issue, we propose Robust Causal Graph Representation Learning (RCGRL) to learn robust graph representations against confounding effects. RCGRL introduces an active approach to generate instrumental variables under unconditional moment restrictions, which empowers the graph representation learning model to eliminate confounders, thereby capturing discriminative information that is causally related to downstream predictions. We offer theorems and proofs to guarantee the theoretical effectiveness of the proposed approach. Empirically, we conduct extensive experiments on a synthetic dataset and multiple benchmark datasets. The results demonstrate that compared with state-of-the-art methods, RCGRL achieves better prediction performance and generalization ability.