
Algorithms That Learn as They Schedule: A Twin-Model Approach to Modern FJS
12 Sept 2025
This article formalizes DAG-based FJS with position-based learning via a MILP using position variables plus a CP Optimizer model.

Faster as You Go: Cracking Job-Shop Puzzles Where Tasks Learn on the Line
12 Sept 2025
Constructive heuristics jump-start MILP/CP on 12-machine tests, laying ground for future metaheuristics in modern FJS variants.

Add New Items, Not New Weights - Personalize Detection Without Backprop
10 Sept 2025
Object-conditioned prototypes plus stats beat FT even with 1-shot, 1-sequence data, and add only 0.8% latency to YOLOv8-n inference.

Few-Shot Personalization of YOLOv8 with Object-Conditioned Bags of Instances
10 Sept 2025
Multi-order stats enrich embeddings, avert neural collapse, and boost cross-domain accuracy while supporting lifelong instance updates.

Boost Few-Shot Instance Accuracy 12% Without Fine-Tuning
9 Sept 2025
By enriching prototype features, the method resists neural collapse and transfers across domains with up to 22% relative gains.

Need Precision Plankton Counts? Why YOLOv5 Shines, But YOLOv8 Adapts
9 Sept 2025
YOLOv5 hits 97% precision on zooplankton; YOLOv8’s DFL handles class imbalance, boosting excrement hits despite scant labels.

Train YOLO in 10 Epochs: A Lean Recipe for Marine Micro-Object Detection
9 Sept 2025
Class imbalance (80% Artemia) and SSIM/MSE checks ensure quality across 50 mg, 100 mg, and control images before YOLO training.

Comparative Study of YOLOv5 and YOLOv8 for Challenging Aquatic Object Detection
9 Sept 2025
Tests in blurry microfluidic images show YOLOv8 more adaptable, while YOLOv5 may need task-specific tweaks for hard marine debris.

A New Approach to 3D Scene Understanding: Replacing Heavy Segmentation Models for a 16x Speedup
26 Aug 2025
This research introduces Open-YOLO 3D, a novel method using 2D object detectors for high-speed, open-vocabulary 3D instance segmentation.