Shielding Collaborative Learning: Mitigating Poisoning Attacks Through Client-Side Detection