Rasa Khosrowshahli

Rasa Khosrowshahli, Stephen Asobiela, Beatrice Ombuki-Berman et al.

Machine unlearning aims to remove the influence of specific training samples while preserving the model's utility. Existing single-objective approaches, such as gradient ascent or random relabeling, often induce catastrophic forgetting due to conflicting optimization dynamics and unbounded forgetting objectives that cause the model to drift from its pre-trained knowledge. We propose Reference-Aligned UnLearning (RAUL), a multi-objective framework that jointly optimizes forgetting and retention by replacing unbounded loss maximization with a bounded KL alignment of predictions on forgotten samples toward a reference distribution representing unseen data, instantiated either as a uniform distribution or an empirical distribution from a held-out reference set, which constrains the forgetting objective and reduces gradient conflict with retention. The resulting multi-objective optimization (MOO) problem is solved via Jacobian descent, which aggregates multiple gradients into a direction that does not conflict. Our results demonstrate that RAUL achieves the closest gap compared to full retraining.

Rasa Khosrowshahli, Shahryar Rahnamayan, Beatrice Ombuki-Berman

Deep learning is mainly based on utilizing gradient-based optimization for training Deep Neural Network (DNN) models. Although robust and widely used, gradient-based optimization algorithms are prone to getting stuck in local minima. In this modern deep learning era, the state-of-the-art DNN models have millions and billions of parameters, including weights and biases, making them huge-scale optimization problems in terms of search space. Tuning a huge number of parameters is a challenging task that causes vanishing/exploding gradients and overfitting; likewise, utilized loss functions do not exactly represent our targeted performance metrics. A practical solution to exploring large and complex solution space is meta-heuristic algorithms. Since DNNs exceed thousands and millions of parameters, even robust meta-heuristic algorithms, such as Differential Evolution, struggle to efficiently explore and converge in such huge-dimensional search spaces, leading to very slow convergence and high memory demand. To tackle the mentioned curse of dimensionality, the concept of blocking was recently proposed as a technique that reduces the search space dimensions by grouping them into blocks. In this study, we aim to introduce Histogram-based Blocking Differential Evolution (HBDE), a novel approach that hybridizes gradient-based and gradient-free algorithms to optimize parameters. Experimental results demonstrated that the HBDE could reduce the parameters in the ResNet-18 model from 11M to 3K during the training/optimizing phase by metaheuristics, namely, the proposed HBDE, which outperforms baseline gradient-based and parent gradient-free DE algorithms evaluated on CIFAR-10 and CIFAR-100 datasets showcasing its effectiveness with reduced computational demands for the very first time.

Rasa Khosrowshahli, Shahryar Rahnamayan, Beatrice Ombuki-Berman

Deep neural networks suffer from storing millions and billions of weights in memory post-training, making challenging memory-intensive models to deploy on embedded devices. The weight-sharing technique is one of the popular compression approaches that use fewer weight values and share across specific connections in the network. In this paper, we propose a multi-objective evolutionary algorithm (MOEA) based compression framework independent of neural network architecture, dimension, task, and dataset. We use uniformly sized bins to quantize network weights into a single codebook (lookup table) for efficient weight representation. Using MOEA, we search for Pareto optimal $k$ bins by optimizing two objectives. Then, we apply the iterative merge technique to non-dominated Pareto frontier solutions by combining neighboring bins without degrading performance to decrease the number of bins and increase the compression ratio. Our approach is model- and layer-independent, meaning the weights are mixed in the clusters from any layer, and the uniform quantization method used in this work has $O(N)$ complexity instead of non-uniform quantization methods such as k-means with $O(Nkt)$ complexity. In addition, we use the center of clusters as the shared weight values instead of retraining shared weights, which is computationally expensive. The advantage of using evolutionary multi-objective optimization is that it can obtain non-dominated Pareto frontier solutions with respect to performance and shared weights. The experimental results show that we can reduce the neural network memory by $13.72 \sim14.98 \times$ on CIFAR-10, $11.61 \sim 12.99\times$ on CIFAR-100, and $7.44 \sim 8.58\times$ on ImageNet showcasing the effectiveness of the proposed deep neural network compression framework.