Knowledge distillation from Large Language Models (LLMs) to locally hosted Small Language Models (SLMs) provides advantages for Data Analysis Code Generation (DACG) such as privacy protection. However, achieving effective distillation without resource-intensive training is challenging. This paper investigates whether LLMs can distill knowledge to SLMs through In-Context Learning (ICL), a training-free method for rapid task adaptation. We present the DARGO: Distillation and Adaptive Reasoning-Guided Orchestration framework, which facilitates automatic knowledge distillation from LLMs to SLMs. DARGO consists of three phases: exploration through an Model Orchestration Interface (MOI), Memory Collection of successful trajectories, and Knoweldge-driven Inference. We evaluate DARGO on three challenging DACG benchmarks (WIKITQ, TABMWP, and BIRD-SQL), each with in-domain training sets that enable detailed analysis of knowledge distillation effectiveness. DARGO demonstrates a substantial relative performance improvement of 27.5% on average for the student SLMs. To further observe generalization capabilities, we evaluate the DARGO across different teacher-student model combinations, knowledge transfer scenarios, and unified memory approaches for more advanced, test-only data analysis tasks. Our findings contribute a novel perspective on distillation methods that enhance performance for SLMs while avoiding intensive fine-tuning.

Large language models (LLMs) have achieved significant performance progress in various natural language processing applications. However, LLMs still struggle to meet the strict requirements for accuracy and reliability in the medical field and face many challenges in clinical applications. Existing clinical diagnostic evaluation benchmarks for evaluating medical agents powered by LLMs have severe limitations. Firstly, most existing medical evaluation benchmarks face the risk of data leakage or contamination.

To investigate this further, we first observe that Claude-3.7-Sonnet Figure 1 shows the average pass rate under budgets of 12,000, 10 14,000, 16,000, and 17,000 tokens. As the data demonstrate, enlarging the thinking budget yields no 11 appreciable improvement in performance. This finding underscores 14 the challenging nature of ENGDESIGN and suggests its value as a rigorous testbed for future efforts 15 to enhance LLMs' engineering design proficiency. Figure 1: Average pass rate (%) of Claude-3.7-Thinking

Many decision in clinical science and epidemiology -- estimating probability of technical success for a clinical trial, assessing comparative effectiveness of two therapies, imputing a placebo effect onto natural history data -- rely on combining sources of information about a clinical cohort that comes from different kinds of studies. Specifically we contrast patient-level sources that provide granular pictures of individual disease course (clinical trial, registries, or electronic health records) with aggregate sources such as published clinical trial results and the TFLs (tables figures and listings). One strategy for combining aggregate with patient-level data sources is to bring each into a common format for a unified analysis. If one wants to maintain the analytic flexibility of patient-level data, then a natural solution is to convert the aggregate data information into a simulated patient-level dataset that recapitulate those aggregate statistics. This is an under-determined inverse problem in that there are many such datasets, and it cannot be well specified without further constraints. FRESH(Fusion of Recent Evidence with Subject Histories) provides a well-defined method for solving this problem, and therefore providing maximal analytic flexibility.

The source code of Minigrid and Miniworld can be found at https://github.com/ To run the experiments, we have implemented the following functionalities: 1. implemented the human trajectory saving for MiniGrid-FourRooms-v0 (copied the ManualControlclass from Minigrid and added 38 lines of code, which are mostly calling data saving functions); 2. implemented the human trajectory saving for MiniWorld-FourRooms-v0 (copied the ManualControlclass from Miniworld and added 45 lines of code, which are mostly calling data saving functions); 3. implemented data saving and plotting for MiniGrid-FourRooms-v0 (33 lines of code, mostly for Matplotlib); 4. implemented data saving and plotting for MiniWorld-FourRooms-v0 (33 lines of code, mostly for Matplotlib). In total, the implementation of this new functionality required 149 lines of code. The source code is hosted on GitHub. We bear all the responsibility in case of violation of rights.

We proceed to show the sparsistency510 of the estimated parameters. First, suppose that Θ t;ij 6= 0 for some time tand index (i,j). Due to 0 < γ < 1, the above inequality implies that bΘt;ij = 0521 for every t and (i,j) 6 St, and bΘt;ij bΘt 1;ij = 0 for every t > 0 and (i,j) 6 Dt. The proof is inspired527 by Corollary 1 in [47]. First, we present the following key lemmas.528

Despite the remarkable success of transformer-based models in various real-world tasks, their underlying mechanisms remain poorly understood. Recent studies have suggested that transformers can implement gradient descent as an in-context learner for linear regression problems and have developed various theoretical analyses accordingly. However, these works mostly focus on the expressive power of transformers by designing specific parameter constructions, lacking a comprehensive understanding of their inherent working mechanisms post-training. In this study, we consider a sparse linear regression problem and investigate how a trained multi-head transformer performs in-context learning. We experimentally discover that the utilization of multi-heads exhibits different patterns across layers: multiple heads are utilized and essential in the first layer, while usually only a single head is sufficient for subsequent layers. We provide a theoretical explanation for this observation: the first layer preprocesses the context data, and the following layers execute simple optimization steps based on the preprocessed context. Moreover, we demonstrate that such a preprocess-then-optimize algorithm can significantly outperform naive gradient descent and ridge regression algorithms. Further experimental results support our explanations. Our findings offer insights into the benefits of multi-head attention and contribute to understanding the more intricate mechanisms hidden within trained transformers.

Data heterogeneity among Federated Learning (FL) users poses a significant challenge, resulting in reduced global model performance. The community has designed various techniques to tackle this issue, among which Knowledge Distillation (KD)-based techniques are common. While these techniques effectively improve performance under high heterogeneity, they inadvertently cause higher accuracy degradation under model poisoning attacks (known as \emph{attack amplification}). This paper presents a case study to reveal this critical vulnerability in KD-based FL systems. We show why KD causes this issue through empirical evidence and use it as motivation to design a hybrid distillation technique. We introduce a novel algorithm, Hybrid Knowledge Distillation for Robust and Accurate FL (HYDRA-FL), which reduces the impact of attacks in attack scenarios by offloading some of the KD loss to a shallow layer via an auxiliary classifier. We model HYDRA-FL as a generic framework and adapt it to two KD-based FL algorithms, FedNTD and MOON. Using these two as case studies, we demonstrate that our technique outperforms baselines in attack settings while maintaining comparable performance in benign settings.