Detection of malicious agents in social learning