Hash Bit Selection via Collaborative Neurodynamic Optimization With Discrete Hopfield Networks.

Area of Science:

• Computer Science
• Machine Learning
• Optimization

Background:

• Hash bit selection (HBS) is crucial for identifying discriminative hash bits from generated hash pools.
• Traditional HBS is often formulated as a complex binary quadratic programming problem.
• Existing methods face challenges in efficiently and effectively selecting optimal hash bits.

Purpose of the Study:

• To reformulate the hash bit selection problem into a quadratic unconstrained binary optimization (QUBO) problem.
• To solve the QUBO problem using a novel collaborative neurodynamic optimization (CNO) approach.
• To evaluate the effectiveness and superiority of the CNO method for HBS.

Main Methods:

• Hash bit selection (HBS) problem reformulated as a quadratic unconstrained binary optimization (QUBO) problem.
• Augmentation of the objective function with a penalty function to achieve the QUBO formulation.
• Solution using collaborative neurodynamic optimization (CNO) with a population of discrete Hopfield networks.
• Determination of key CNO hyperparameters via Monte Carlo simulations.

Main Results:

• The reformulated QUBO problem was effectively solved using the CNO approach.
• Experimental results on three benchmark datasets validated the CNO method's performance.
• The CNO approach demonstrated superior performance compared to several existing HBS methods.

Conclusions:

• The proposed reformulation of HBS as a QUBO problem is effective.
• Collaborative neurodynamic optimization (CNO) offers a powerful and efficient solution for HBS.
• The CNO method significantly advances the state-of-the-art in hash bit selection.