Humans break down continuous speech into meaningful units by integrating multiple cues whose importance and relative weight change throughout development. We investigate whether segmentation abilities differ between younger and older adults while exploring how age of acquisition, proficiency, language use, and frequency of language switch determine Basque-Spanish younger and older bilinguals' use of lexico-semantic, morphosyntactic, and phonological information in L2 word segmentation. Younger and older L1Spanish-L2Basque bilinguals (n = 158) varying across the four dimensions of interest and a control group of L1Basque bilinguals (n = 63) were tested on a phoneme localization task in Basque. Age did not affect word segmentation in either group. Proficiency modulated performance of the L2Basque bilinguals, with the greatest accuracy gains in sentences containing all sources of information, followed by sentences containing only morphosyntactic and phonological cues, then sentences comprising only phonological information. Frequency of language switch and fluid intelligence were associated with better overall performance, though their contribution to explaining the variance was very small. Finally, natives benefited more than non-natives from lexico-semantic cues, and from morphosyntactic and phonological cues for word-initial targets. These findings suggest the stability of segmentation abilities across the adult lifespan, and evidence how different bilingual factors shape multiple cue use in L2 segmentation.
With the rapid proliferation of large language models and vision-language models, AI agents have evolved from isolated, task-specific systems into autonomous, interactive entities capable of perceiving, reasoning, and acting without human intervention. As these agents proliferate across virtual and physical environments, from virtual assistants to embodied robots, the need for a unified, agent-centric infrastructure becomes paramount. In this survey, we introduce the Internet of Agents (IoA) as a foundational framework that enables seamless interconnection, dynamic discovery, and collaborative orchestration among heterogeneous agents at scale. We begin by presenting a general IoA architecture, highlighting its hierarchical organization, distinguishing features relative to the traditional Internet, and emerging applications. Next, we analyze the key operational enablers of IoA, including capability notification and discovery, adaptive communication protocols, dynamic task matching, consensus and conflict-resolution mechanisms, and incentive models. Finally, we identify open research directions toward building resilient and trustworthy IoA ecosystems.
This work presents a method for implementing dynamic difficulty adjustment in the arcade game of Air Hockey using reinforcement learning. The resulting AI-controlled opponent is capable of adapting its skill level to the player's performance to maintain engagement and provide a balanced gameplay experience. The approach relies on generating several AI agents through progressively longer training durations, resulting in distinct and smoothly transitioning difficulty levels that can be switched dynamically. We discuss how this scheme can be extended with manually selected parameters that influence physical aspects of the agent's behavior-such as movement speed, reaction latency, and control precision-to complement the variations in decision-making quality. The proposed method is applicable to a wide range of video games, and experimental results demonstrate its effectiveness in producing adaptive and varied opponent behavior.
Introduction cervical cancer ranks as the fourth most common cancer among females worldwide. Approximately 528,000 new cases of cervical cancer are reported annually, and about 85% of them occur in less-developed countries. The lack of skilled medical staff and pre-screening procedures is the main cause of the high fatality rate in these countries. Cervicography images are the gold standard procedure for the evaluation of cervical cancer; however, the high intra-class inconsistency makes the diagnosis process more challenging for skilled medical specialists.Method In this work, we propose a fully automated computer-aided diagnosis (CAD) system for classifying cervical cancer using Cervicography images. Data augmentation is performed in the initial phase to address dataset imbalance. Subsequently, we proposed two novel deep learning modules: the 11-Parallel Inverted Residual Bottleneck Blocks (11-PIRBnet) architecture and the 9-Parallel Inverted Residual blocks with Self-Attention Mechanism (9-PIRSANet). Both modules are fused at the network level via a depth concatenation layer to form a new network, 375NFNet. The proposed network is trained on the selected dataset, whereas the hyperparameters are initialized through Bayesian Optimization (BO). For feature extraction, a depth concatenation layer is used during testing to combine information from both deep learning modules. Finally, the extracted features are classified using a shallow neural network (SNN) to produce the final classification.Result To evaluate the model, experiments were conducted on a publicly available cervical screening dataset of Cervicography images, and results demonstrate an accuracy of 95.5%, a precision of 95.4%, and an area under the curve of 0.97. When compared with several pre-trained techniques, the proposed architecture achieved significant improvement in accuracy, precision, and number of trainable parameters.Conclusion The proposed 375NFNet architecture demonstrates remarkable accuracy and efficiency in classifying cervical cancer through cervicography images, which shows its potential as a valuable tool in resource-constrained environments.
Large Language Models (LLMs) are widely deployed in applications that accept user-submitted content, such as uploaded documents or pasted text, for tasks like summarization and question answering. In this paper, we identify a new class of attacks, prompt in content injection, where adversarial instructions are embedded in seemingly benign inputs. When processed by the LLM, these hidden prompts can manipulate outputs without user awareness or system compromise, leading to biased summaries, fabricated claims, or misleading suggestions. We demonstrate the feasibility of such attacks across popular platforms, analyze their root causes including prompt concatenation and insufficient input isolation, and discuss mitigation strategies. Our findings reveal a subtle yet practical threat in real-world LLM workflows.
