Laksh Patel

Second-Order State Hallucinations (SOSH) is a novel methodology for mitigating attacks in formation control of multi-agent systems. Traditional multi-agent systems, when an error occurs, experience cascading faults throughout the network. SOSH uses residual analysis to detect faulty agents within a threshold and “hallucinate” replacement states for the compromised nodes. It then updates the network topology to exclude those nodes. SOSH estimates each attacked node’s position using both velocity and acceleration (second-order approximation), enabling practical deployment in search-and-rescue, platooning, traffic control, and military applications.

MITRA is a research project in the Berkeley AI Research lab in EECS at UC Berkeley. Leveraging a robust corpus of over four million sentence pairs from various sources and using Google's MADLAD-400 model as a foundation, MITRA has fine-tuned a specialized translation model that enhances fluency and significantly expands access to ancient wisdom texts.

This is a data-centric framework that assigns each pretraining sample a difficulty score (early-epoch loss) and a memorization score (frequency of “forget events”), then partitions examples into four quadrants to guide targeted pruning and up-/down-weighting.

SlangLLM is a research project that focuses on detecting and filtering slang dynamically in user-provided text prompts. The system combines natural language processing (NLP), semantic similarity analysis, and toxicity classification to enhance safe communication and mitigate harm in interactions with large language models.

Designed as a scalable, resilient solution for decentralized flow allocation in modern supply-chain networks, it enables autonomous warehouses, carriers, and retailers to coordinate optimally despite high latency, packet loss, and dynamically changing network conditions.

This project uses neuronal modeling tools such as SNNAP and Yale's NEURON software to build biologically accurate networks that mimic classical computing gates. It introduces the Yield Neuron Circuit and the Clone Neuron Circuit, enabling buffers and signal duplication in artificial neural networks.