Large language models (LLMs) are increasingly deployed in security-sensitive applications, where they must follow system-or developer-specified instructions that define the intended task behavior, while completing benign user requests.When adversarial instructions appear in user queries or externally retrieved content, models may override intended logic.Recent defenses rely on supervised finetuning with benign and malicious labels.Although these methods achieve high attack rejection rates, we find that they rely on narrow correlations in defense data rather than harmful intent, leading to systematic rejection of safe inputs.We analyze three recurring shortcut behaviors induced by defense fine-tuning.Position bias arises when benign content placed later in a prompt is rejected at much higher rates; across reasoning benchmarks, suffix-task rejection rises from below 10% to as high as 90%.Token trigger bias occurs when strings common in attack data raise rejection probability even in benign contexts; inserting a single trigger token increases false refusals by up to 50%.Topic generalization bias reflects poor generalization beyond the defense data distribution, with defended models suffering test-time accuracy drops of up to 40%.These findings suggest that current prompt-injection defenses frequently respond to attack-like surface patterns rather than the underlying intent.We introduce controlled diagnostic datasets and a systematic evaluation across two base models and multiple defense pipelines, highlighting limitations of supervised fine-tuning for reliable LLM security.

''Grokking'' is a phenomenon where a neural network first memorizes training data and generalizes poorly, but then suddenly transitions to near-perfect generalization after prolonged training. While intriguing, this delayed generalization phenomenon compromises predictability and efficiency. Ideally, models should generalize directly without delay. To this end, this paper proposes GrokTransfer, a simple and principled method for accelerating grokking in training neural networks, based on the key observation that data embedding plays a crucial role in determining whether generalization is delayed. GrokTransfer first trains a smaller, weaker model to reach a nontrivial (but far from optimal) test performance. Then, the learned input embedding from this weaker model is extracted and used to initialize the embedding in the target, stronger model. We rigorously prove that, on a synthetic XOR task where delayed generalization always occurs in normal training, GrokTransfer enables the target model to generalize directly without delay. Moreover, we demonstrate that, across empirical studies of different tasks, GrokTransfer effectively reshapes the training dynamics and eliminates delayed generalization, for both fully-connected neural networks and Transformers.

Graph neural networks (GNNs) have displayed significant potential in various graph-based learning tasks. However, the computational demands of deploying GNNs on large-scale graphs can grow exponentially. A recent method, termed unified graph sparsification (UGS), shows that there exists a pair consisting of a subgraph and a sparse subnetwork, called graph lottery ticket (GLT), that can effectively speed up GNN inference. However, despite their advantages, the performance of GLTs against adversarial structure perturbations remains largely unexplored. In this paper, we investigate the resilience of GLTs against different structure perturbation attacks under the poisoning attack setting. The evaluation results show that the GLTs identified by UGS are vulnerable and exhibit a large drop in classification accuracy for the adversarially perturbed graphs. We then propose a new technique for defending UGS that leverages self-training to find GLTs that are more resilient and can achieve better performance than plain UGS.

In this paper, a novel medical knowledge graph in Chinese approach applied in smart healthcare based on IoT and WoT is presented, using deep neural networks combined with self‐attention to generate medical knowledge graph to make it more convenient for performing disease diagnosis and providing treatment advisement. Although great success has been made in the medical knowledge graph in recent studies, the issue of comprehensive medical knowledge graph in Chinese appropriate for telemedicine or mobile devices have been ignored. In our study, it is a working theory which is based on semantic mobile computing and deep learning. When several experiments have been carried out, it is demonstrated that it has better performance in generating various types of medical knowledge graph in Chinese, which is similar to that of the state‐of‐the‐art. Also, it works well in the accuracy and comprehensive, which is much higher and highly consisted with the predictions of the theoretical model. It proves to be inspiring and encouraging that our work involving studies of medical knowledge graph in Chinese, which can stimulate the smart healthcare development.