Autoformalization aims to convert informal mathematical proofs into machine-verifiable formats, bridging the gap between natural and formal languages. However, ensuring semantic alignment between the informal and formalized statements remains challenging. Existing approaches heavily rely on manual verification, hindering scalability. To address this, we introduce \textsc{FormalAlign}, the first automated framework designed for evaluating the alignment between natural and formal languages in autoformalization. \textsc{FormalAlign} trains on both the autoformalization sequence generation task and the representational alignment between input and output, employing a dual loss that combines a pair of mutually enhancing autoformalization and alignment tasks. Evaluated across four benchmarks augmented by our proposed misalignment strategies, \textsc{FormalAlign} demonstrates superior performance. In our experiments, \textsc{FormalAlign} outperforms GPT-4, achieving an Alignment-Selection Score 11.58\% higher on \forml-Basic (99.21\% vs. 88.91\%) and 3.19\% higher on MiniF2F-Valid (66.39\% vs. 64.34\%). This effective alignment evaluation significantly reduces the need for manual verification. Both the dataset and code can be accessed via~\url{https://github.com/rookie-joe/FormalAlign}.

Crafting effective features is a crucial yet labor-intensive and domain-specific task within machine learning pipelines. Fortunately, recent advancements in Large Language Models (LLMs) have shown promise in automating various data science tasks, including feature engineering. But despite this potential, evaluations thus far are primarily based on the end performance of a complete ML pipeline, providing limited insight into precisely how LLMs behave relative to human experts in feature engineering. To address this gap, we propose ELF-Gym, a framework for Evaluating LLM-generated Features. We curated a new dataset from historical Kaggle competitions, including 251 "golden" features used by top-performing teams. ELF-Gym then quantitatively evaluates LLM-generated features by measuring their impact on downstream model performance as well as their alignment with expert-crafted features through semantic and functional similarity assessments. This approach provides a more comprehensive evaluation of disparities between LLMs and human experts, while offering valuable insights into specific areas where LLMs may have room for improvement. For example, using ELF-Gym we empirically demonstrate that, in the best-case scenario, LLMs can semantically capture approximately 56% of the golden features, but at the more demanding implementation level this overlap drops to 13%. Moreover, in other cases LLMs may fail completely, particularly on datasets that require complex features, indicating broad potential pathways for improvement.

However, the questions of why and when it is better than other algorithms in few-shot classification remain to be explored. In this paper, we perform pre-experiments by adjusting the proportion of label noise and the degree of task heterogeneity in the dataset. We use the metric of Singular Vector Canonical Correlation Analysis to quantify the representation stability of the neural network and thus to compare the behavior of meta-learning and classical learning algorithms. We find that benefiting from the bi-level optimization strategy, the meta-learning algorithm has better robustness to label noise and heterogeneous tasks. Based on the above conclusion, we argue a promising future for meta-learning in the unsupervised area, and thus propose DHM-UHT, a dynamic head meta-learning algorithm with unsupervised heterogeneous task construction. The core idea of DHM-UHT is to use DBSCAN and dynamic head to achieve heterogeneous task construction and meta-learn the whole process of unsupervised heterogeneous task construction. As an example, the optimization-based meta-learning algorithm Finn et al. (2017); Raghu et al. (2020); Nichol et al. (2018) has been shown to demonstrate excellent generalization performance in few-shot learning and reinforcement learning. In these areas, the more commonly used pre-train and fine-tune strategy exhibits disadvantages regarding training overhead, reliance on massive samples, and accuracy. However, in recent years, new research has shown that models pre-trained by the classical Whole-Class Training (WCT) strategy exhibit comparable or even better accuracy on multiple few-shot image classification datasets Tian et al. (2020); Chen et al. (2021). The inconsistent conclusions described above confuse us about the nature of meta-learning, and in turn hinder us from developing the area.

We prove that all 'gradient span algorithms' have asymptotically deterministic behavior on scaled Gaussian random functions as the dimension tends to infinity. In particular, this result explains the counterintuitive phenomenon that different training runs of many large machine learning models result in approximately equal cost curves despite random initialization on a complicated non-convex landscape. The distributional assumption of (non-stationary) isotropic Gaussian random functions we use is sufficiently general to serve as realistic model for machine learning training but also encompass spin glasses and random quadratic functions.

As digital scamming explodes in Southeast Asia, including so called "pig butchering" investment scams, the United Nations Office on Drugs and Crime (UNODC) issued a comprehensive report this week with a dire warning about the rapid growth of this criminal ecosystem. Many digital scams have traditionally relied on social engineering, or tricking victims into giving away their money willingly, rather than leaning on malware or other highly technical methods. But researchers have increasingly sounded the alarm that scammers are incorporating generative AI content and deepfakes to expand the scale and effectiveness of their operations. And the UN report offers the clearest evidence yet that these high tech tools are turning an already urgent situation into a crisis. In addition to buying written scripts to use with potential victims or relying on templates for malicious websites, attackers have increasingly been leaning on generative AI platforms to create communication content in multiple languages and deepfake generators that can create photos or even video of nonexistent people to show victims and enhance verisimilitude.

Food security (FS) is a complex and multifaceted problem, influenced by several factors such as weather events, economic shocks, and natural disasters. Understanding the dynamics of food security is crucial for effective policymaking and humanitarian efforts. While conflicts and violent events increasingly stand out as key drivers of food crises[1], the depth of their impact remains largely underexplored. Examining the quantitative aspects of this impact is essential for developing more targeted interventions and strategies to address the complex interplay between conflict and food security. Existing research tends to be qualitative in nature (Kemmerling et al.2022; Brown et al. 2020; Brown et al. 2021), leaving a significant gap in understanding the quantitative aspects of how conflicts impact FS levels. By delving into quantitative analyses, we can not only enhance our comprehension of the magnitude of the problem but also pave the way for evidence-based decision-making in efforts to alleviate food insecurity in conflict-affected regions. Regarding the qualitative study of conflicts and FS, Kemmerling et al.(2022)[2] provided a comprehensive explanation on how violence and armed conflicts impact FS through destruction, displacement, financing of conflicts and food being used as a weapon. The authors call for better conflict data collection, and an increase in focus on the study of conflicts early warnings.

Large language models (LLMs) sometimes demonstrate poor performance on knowledge-intensive tasks, commonsense reasoning is one of them. Researchers typically address these issues by retrieving related knowledge from knowledge graphs or employing self-enhancement methods to elicit knowledge in LLMs. However, noisy knowledge and invalid reasoning issues hamper their ability to answer questions accurately. To this end, we propose a novel method named eliciting, filtering and integrating knowledge in large language model (LINKED). In it, we design a reward model to filter out the noisy knowledge and take the marginal consistent reasoning module to reduce invalid reasoning. With our comprehensive experiments on two complex commonsense reasoning benchmarks, our method outperforms SOTA baselines (up to 9.0% improvement of accuracy). Besides, to measure the positive and negative impact of the injected knowledge, we propose a new metric called effectiveness-preservation score for the knowledge enhancement works. Finally, through extensive experiments, we conduct an in-depth analysis and find many meaningful conclusions about LLMs in commonsense reasoning tasks.

The paper evaluates the USV's autonomy through open-water

Iterative refinement has emerged as an effective paradigm for enhancing the capabilities of large language models (LLMs) on complex tasks. However, existing approaches typically implement iterative refinement at the application or prompting level, relying on autoregressive (AR) modeling. The sequential token generation in AR models can lead to high inference latency. To overcome these challenges, we propose Context-Wise Order-Agnostic Language Modeling (COrAL), which incorporates iterative refinement directly into the LLM architecture while maintaining computational efficiency. Our approach models multiple token dependencies within manageable context windows, enabling the model to perform iterative refinement internally during the generation process. Leveraging the order-agnostic nature of COrAL, we introduce sliding blockwise order-agnostic decoding, which performs multi-token forward prediction and backward reconstruction within context windows. This allows the model to iteratively refine its outputs in parallel in the sliding block, effectively capturing diverse dependencies without the high inference cost of sequential generation. Empirical evaluations on reasoning tasks demonstrate that COrAL improves performance and inference speed, respectively, achieving absolute accuracy gains of $4.6\%$ on GSM8K and $4.0\%$ on LogiQA, along with inference speedups of up to $3.9\times$ over next-token baselines. Preliminary results on code generation indicate a drop in pass rates due to inconsistencies in order-agnostic outputs, highlighting the inherent quality--speed trade-off. Our code is publicly available at https://github.com/YuxiXie/COrAL.

Inductive reasoning is an essential capability for large language models (LLMs) to achieve higher intelligence, which requires the model to generalize rules from observed facts and then apply them to unseen examples. We present {\scshape Mirage}, a synthetic dataset that addresses the limitations of previous work, specifically the lack of comprehensive evaluation and flexible test data. In it, we evaluate LLMs' capabilities in both the inductive and deductive stages, allowing for flexible variation in input distribution, task scenario, and task difficulty to analyze the factors influencing LLMs' inductive reasoning. Based on these multi-faceted evaluations, we demonstrate that the LLM is a poor rule-based reasoner. In many cases, when conducting inductive reasoning, they do not rely on a correct rule to answer the unseen case. From the perspectives of different prompting methods, observation numbers, and task forms, models tend to consistently conduct correct deduction without correct inductive rules. Besides, we find that LLMs are good neighbor-based reasoners. In the inductive reasoning process, the model tends to focus on observed facts that are close to the current test example in feature space. By leveraging these similar examples, the model maintains strong inductive capabilities within a localized region, significantly improving its deductive performance.