While critical for alignment, Supervised Fine-Tuning (SFT) incurs the risk of catastrophic forgetting, yet the layer-wise emergence of instruction-following capabilities remains elusive.We investigate this mechanism via a comprehensive analysis utilizing informationtheoretic, geometric, and optimization metrics across model scales (1B-32B).Our experiments reveal a distinct depth-dependent pattern: middle layers (20%-80%) are stable, whereas final layers exhibit high sensitivity.Leveraging this insight, we propose Mid-Block Efficient Tuning, which selectively updates these critical intermediate layers.Empirically, our method outperforms standard LoRA up to 10.2% on GSM8K (OLMo2-7B) with reduced parameter overhead, demonstrating that effective alignment is architecturally localized rather than distributed.The code is publicly available at https://github.com/lshowway/base.

Graph convolutional networks (GCNs) are widely used in graph-based applications, such as social networks and recommendation systems. Nevertheless, large-scale graphs or deep aggregation layers in full-batch GCNs consume significant GPU memory, causing out-of-memory (OOM) errors on mainstream GPUs (e.g., 29-GB memory consumption on the Ogbn-products graph with five layers). The subgraph sampling methods reduce memory consumption to achieve lightweight GCNs by partitioning the graph into multiple subgraphs and sequentially training GCNs on each subgraph. However, these methods yield gaps among subgraphs, i.e., GCNs can only be trained based on subgraphs instead of global graph information, which reduces the accuracy of GCNs. In this article, we propose PromptGCN, a novel prompt-based lightweight GCN model to bridge the gaps among subgraphs. First, the learnable prompt embeddings are designed to obtain global information. Then, the prompts are attached to each subgraph to transfer the global information among subgraphs. Extensive experimental results on seven large-scale graphs demonstrate that PromptGCN exhibits superior performance compared to baselines. Notably, PromptGCN improves the accuracy of subgraph sampling methods by up to 5.48% on the Flickr dataset. Overall, PromptGCN is easily integrable with any subgraph sampling method to obtain a lightweight GCN model with higher accuracy.

Over the past two decades, frequent blackouts have highlighted the critical importance of ensuring the security of power grid. The integration of cyber and physical domains through intelligent devices has increased the risk of asynchronous attacks that can trigger cascading failures. One effective approach to mitigating this threat is to identify vulnerable sequences, which are sequences of transmission lines that can cause large-scale failures in the power grid. This article proposes an event-triggered hybrid system model to characterize the generation and propagation mechanism of sequential cascading failures. In addition, the problem of identifying vulnerable sequences is formulated as a Markov decision process. To solve the sequential decision problem in approximately contiguous states, a vulnerable sequence identification method based on reinforcement learning is designed. Furthermore, a topological feature embedding algorithm based on matrix decomposition is proposed to improve identification performance. To evaluate the effectiveness of the proposed method, various numerical experiments are conducted on IEEE 30-bus and ACTIVSg 200-bus systems. The results of these experiments demonstrate the excellent performance of the proposed method.

A fitness assignment process transforms the features (such as the objective value) of a candidate solution to a scalar fitness, which then is the basis for selection. Under frequency fitness assignment (FFA), the fitness corresponding to an objective value is its encounter frequency in selection steps and is subject to minimization. FFA creates algorithms that are not biased toward better solutions and are invariant under all injective transformations of the objective function value. We investigate the impact of FFA on the performance of two theory inspired, state-of-the-art evolutionary algorithms, the Greedy (2+1) GA and the self-adjusting$(1+(\lambda,\lambda))$GA. FFA improves their performance significantly on some problems that are hard for them. In our experiments, one FFA-based algorithm exhibited mean runtimes that appear to be polynomial on the theory-based benchmark problems in our study, including traps, jumps, and plateaus. We propose two hybrid approaches that use both direct and FFA-based optimization and find that they perform well. All FFA-based algorithms also perform better on satisfiability problems than any of the pure algorithm variants.

Under frequency fitness assignment (FFA), the fitness corresponding to an objective value is its encounter frequency in fitness assignment steps and is subject to minimization. FFA renders optimization processes invariant under bijective transformations of the objective function value. On TwoMax, Jump, and Trap functions of dimension s, the classical (1 + 1)-EA with standard mutation at rate 1/s can have expected runtimes exponential in s. In our experiments, a (1 + 1)-FEA, the same algorithm but using FFA, exhibits mean runtimes that seem to scale as s2ln s. Since Jump and Trap are bijective transformations of OneMax, it behaves identical on all three. On OneMax, LeadingOnes, and Plateau problems, it seems to be slower than the (1 + 1)-EA by a factor linear in s. The (1 + 1)-FEA performs much better than the (1 + 1)-EA on W-Model and MaxSat instances. We further verify the bijection invariance by applying the Md5 checksum computation as transformation to some of the above problems and yield the same behaviors. Finally, we show that FFA can improve the performance of a memetic algorithm for job shop scheduling.

Landscape freeze/thaw (FT) state is a key variable in Earth's carbon cycle. NASA's Soil Moisture Active Passive (SMAP) satellite mission, launched in January 2015, provides global retrievals of FT state every two to three days. Validating SMAP FT observations with in-situ observations is difficult due to the substantial scale mismatch between a point estimate and a satellite footprint, inducing “representativeness errors” in the in-situ observations. Triple collocation (TC) is a validation technique that addresses this problem by combining estimates from in-situ, model and spaceborne estimates to obtain error estimates for all three products, without assuming that any product is error-free. Unfortunately, it fails when applied to binary or categorical variables, such as landscape FT state. In this study, we use a new variant of TC - categorical triple collocation (CTC) - that can be applied to binary variables, to validate the SMAP FT product across northern land regions (>45N).

In this study, we present a detailed inter-annual analysis of four remotely sensed leaf area index (LAI) products: GLASS, GLOBALBNU, GLOBMAP, and MODIS LAI, and conduct correlation analysis between LAI and three meteorological variables over China. The results manifest that the four products agree well in most of northern regions of China, with higher correlation coefficient and the same response to meteorological factors. The phenology matches well for different biome types, reaching the peak in July or August. The changing trend of LAI from 2001-2011 is almost same for four products, with approximately 56% greening and 44% browning. However, abrupt changes occur at different time among four products. LAI is positively correlated with meteorological station annual average precipitation and temperature, while negatively correlated with annual average sunshine duration in spatial scale.