Deploying 6G network slices across administrative domains requires an online decision for each arriving request, and this joint problem is NP-hard under continuous arrivals. Experiments on multi-domain topologies show that no fixed planning strategy dominates, which identifies the plan layer as the decisive performance lever.

• Two LLM agents translate service intents into structured requests and generate abstract placement plans.
• A cooperative multi-agent RL coordinator selects the cross-domain partition, treating the plan as a persistent soft prior, while per-domain agents place the functions.
• An outcome-driven episodic memory feeds verified successes and labelled failures back to the planner.
• Admission improves by 4.6–11.5 percentage points over the same planner without memory, on held-out topology families.

Federated fine-tuning of LLMs across edge devices with different resources, without sharing raw data.

• Phase 1: gradient fingerprinting and task clustering.
• Phase 2: device-aware heterogeneous LoRA rank allocation.
• Phase 3: split federated learning.
• Phase 4: Laplacian regularization for personalization.

Small-scale sentiment classification suffers from data scarcity, which limits model generalization. This study systematically compares three text augmentation strategies under a controlled, reproducible framework.

• EDA (Easy Data Augmentation) based on token-level synonym replacement, insertion, deletion, and swapping.
• Back-translation using English↔French MarianMT models to create high-fidelity paraphrases.
• Contextual token substitution (NLPaug-style) with pre-trained language models for semantics-preserving edits.

Methods: Experiments use a 5,000-sample IMDb subset with 100% augmentation, 10-fold cross-validation, and fixed seeds, comparing traditional classifiers (Logistic Regression, Random Forest) and a fine-tuned BERT base model on accuracy, F1, AUC, and effect sizes.

• All augmentation strategies yield significant and statistically robust performance gains over non-augmented baselines.
• Contextual augmentation delivers the most consistent improvements for BERT, reaching about 97% test accuracy on the augmented setting.
• EDA and back-translation provide larger relative gains for traditional models, especially Random Forest, while exhibiting different diversity–cost trade-offs.

Lip reading plays a crucial role in applications such as speech recognition, assistive technologies for the hearing-impaired, and biometric authentication. However, performance often degrades when the speaker varies, the environment changes, or the speech sounds are visually similar. To overcome these challenges, we propose LipLingo:

• Based on the LipNet architecture
• Enhanced with standardized preprocessing for consistent mouth-region representation
• Uses rotational validation for better generalization to unseen speakers
• Adds a flexible similarity evaluation for more reliable sentence-level assessment
• Combines 3D convolutional layers (spatio-temporal features) and bidirectional recurrent layers (sequence modeling)
• Optimized with Adam optimizer and CTC loss

Results on the GRID dataset:

• 95.0% character-level accuracy (CER 4.99%)
• 87.2% sentence-level accuracy (corpus-level WER 13.1%)
• Outperforms baseline LipNet