Adnan Saeed

Adnan Saeed is a deep learning researcher working on medical image analysis, with a focus on multimodal architectures, graph neural networks, and evidential deep learning for clinical imaging tasks. His peer reviewed research has appeared in journals across machine learning and biomedical signal processing. At AI Trend Blend he turns recent papers into clear, practical explainers, with an emphasis on what a method actually does and where it holds up, written for readers who want depth without the hype.

AgentDropoutV2: Test-Time Rectify-or-Reject Pruning for Multi-Agent Systems.

AgentDropoutV2: Test-Time Rectify-or-Reject Pruning for Multi-Agent Systems

1 Comment / Machine Learning, Agent Systems /

AgentDropoutV2: Test-Time Rectify-or-Reject Pruning for Multi-Agent Systems | AI Security Research AISecurity Research Machine Learning About Multi-Agent Systems · arXiv:2602.23258v1 [cs.AI] · 16 min read AgentDropoutV2: Teaching Multi-Agent Systems to Self-Correct Through Test-Time Rectify-or-Reject Pruning A novel test-time framework that intercepts and iteratively rectifies erroneous agent outputs using retrieval-augmented adversarial indicators, achieving 6.3% accuracy improvement […]

AgentDropoutV2: Test-Time Rectify-or-Reject Pruning for Multi-Agent Systems Read More »

ACCF: Adversarial Contrastive Collaborative Filtering.

ACCF: Adversarial Contrastive Collaborative Filtering

Leave a Comment / Machine Learning, Cybersecurity /

ACCF: Adversarial Contrastive Collaborative Filtering | AI Security Research AISecurity Research Machine Learning About Recommender Systems · Knowledge-Based Systems 2026 · 14 min read ACCF: Teaching Recommender Systems to Learn from Adversity Through Contrastive Learning A novel training paradigm that integrates adversarial perturbations with instance-sensitive optimization to enhance robustness and generality in graph neural network-based

ACCF: Adversarial Contrastive Collaborative Filtering Read More »

The FedDRLPD system architecture.

FedDRLPD: Deep Reinforcement Learning Defense Against Poisoning Attacks in Federated Learning

Leave a Comment / Machine Learning, Cybersecurity /

FedDRLPD: Deep Reinforcement Learning Defense Against Poisoning Attacks in Federated Learning | AI Security Research AISecurity Research Machine Learning About Federated Learning Security · Knowledge-Based Systems 2026 · 16 min read FedDRLPD: Teaching AI to Defend Itself Against Poisoning Attacks Through Deep Reinforcement Learning A novel defense framework that integrates Deep Q-Network algorithms with Mahalanobis

FedDRLPD: Deep Reinforcement Learning Defense Against Poisoning Attacks in Federated Learning Read More »

K2-Agent: The Cognitive Architecture That Taught AI to Think Like Humans About Mobile Tasks.

K2-Agent: Co-Evolving Know-What and Know-How for Hierarchical Mobile Device Control

Leave a Comment / Machine Learning, Agent Systems /

K2-Agent: Co-Evolving Know-What and Know-How for Hierarchical Mobile Device Control | AI Security Research AISecurity Research Machine Learning About Agent Systems · ICLR 2026 · 18 min read K2-Agent: The Cognitive Architecture That Taught AI to Think Like Humans About Mobile Tasks A hierarchical framework separates “knowing what” from “knowing how” — enabling co-evolution of

K2-Agent: Co-Evolving Know-What and Know-How for Hierarchical Mobile Device Control Read More »

PDF: PUF-based DNN Fingerprinting for Knowledge Distillation Traceability.

PDF: PUF-based DNN Fingerprinting for Knowledge Distillation Traceability

Leave a Comment / Machine Learning, Cybersecurity /

PDF: PUF-based DNN Fingerprinting for Knowledge Distillation Traceability | AI Security Research AISecurity Research Machine Learning About Model Security · DAC 2026, Long Beach, CA · 15 min read The Hardware Fingerprint That Traces Stolen AI Models Back to Their Source A novel PUF-based framework embeds unclonable hardware signatures into teacher models during knowledge distillation,

PDF: PUF-based DNN Fingerprinting for Knowledge Distillation Traceability Read More »

Preference Score Distillation: Leveraging 2D Rewards to Align Text-to-3D Generation with Human Preference.

Preference Score Distillation: Leveraging 2D Rewards to Align Text-to-3D Generation with Human Preference

Leave a Comment / Machine Learning /

Preference Score Distillation: Leveraging 2D Rewards to Align Text-to-3D Generation with Human Preference | MedAI Research 3D Generation · Computer Vision, 2026 · 18 min read A breakthrough framework called PSD bridges the gap between 2D aesthetic preferences and 3D generation — without requiring a single 3D training sample, by reformulating RLHF as a classifier-free

Preference Score Distillation: Leveraging 2D Rewards to Align Text-to-3D Generation with Human Preference Read More »

TPMRI Framework Architecture.

TPMRI: How Three-Stage Progressive Fusion Is Solving RGB-T Tracking’s Temporal Blindness

Leave a Comment / Machine Learning /

TPMRI: How Three-Stage Progressive Fusion Is Solving RGB-T Tracking’s Temporal Blindness | MedAI Research MedAI Research Machine Learning About Computer Vision · Knowledge-Based Systems, 2026 · 14 min read When RGB-T Trackers Lose Track: How TPMRI Learned to Remember Through Time TPMRI introduces a three-stage progressive fusion framework that fixes RGB-T tracking’s most frustrating failures

TPMRI: How Three-Stage Progressive Fusion Is Solving RGB-T Tracking’s Temporal Blindness Read More »

RCD framework addresses three critical bottlenecks in text-to-image generation.

RCD: How Three Simple Fixes Are Solving Stable Diffusion’s Biggest Problem

Leave a Comment / Machine Learning /

RCD: How Three Simple Fixes Are Solving Stable Diffusion’s Biggest Problem | MedAI Research MedAI Research Machine Learning About Deep Learning · TPAMI, 2026 · 16 min read When Stable Diffusion Forgets: How RCD Learned to Remember Every Detail RCD introduces a training-free framework that fixes text-to-image diffusion models’ most frustrating failures — missing objects

RCD: How Three Simple Fixes Are Solving Stable Diffusion’s Biggest Problem Read More »

The WEMoE framework transforms critical MLP modules into dynamic mixture-of-experts structures while statically merging non-critical components. Input-dependent routing weights allow the model to adaptively blend task-specific knowledge, achieving superior multi-task performance over static merging methods.

WEMoE: How a Mixture-of-Experts Approach Is Solving the Multi-Task Model Merging Problem

Leave a Comment / Machine Learning /

WEMoE: How a Mixture-of-Experts Approach Is Solving the Multi-Task Model Merging Problem | MedAI Research MedAI Research Machine Learning About Deep Learning · TPAMI, 2026 · 18 min read The Static Model Merging Problem — and How WEMoE Learned to Adapt WEMoE introduces a dynamic mixture-of-experts approach to multi-task model merging, transforming how we combine

WEMoE: How a Mixture-of-Experts Approach Is Solving the Multi-Task Model Merging Problem Read More »

the proposed ESM-AnatTractNet model

ESM-AnatTractNet: Deep Learning for Eloquent White Matter Tractography in Pediatric Epilepsy Surgery

Leave a Comment / Machine Learning, Medical AI /

ESM-AnatTractNet: Deep Learning for Eloquent White Matter Tractography in Pediatric Epilepsy Surgery | MedAI Research MedAI Research Machine Learning About Neurosurgical AI · Medical Image Analysis, 2026 · 22 min read The Deep Learning System That Learned to Map Eloquent Brain Circuits from Electrical Stimulation and Anatomy ESM-AnatTractNet integrates electrophysiological validation with anatomical context to

ESM-AnatTractNet: Deep Learning for Eloquent White Matter Tractography in Pediatric Epilepsy Surgery Read More »