Apostolos Avranas

Apostolos Avranas

Modern optimizers combine gradients from the current mini-batch with historical optimization state, such as momentum or adaptive moments. While highly effective, aggregating across the batch and incorporating this history can produce parameter updates that increase the loss of individual samples. We term this effect harm and formalize the parameter update as an optimization problem that explicitly minimizes the conflicting impact of both batch averaging and past optimization state on current data. Because the exact formulation is intractable, we introduce a highly efficient proxy. We first reduce the problem's dimensionality to the batch size, and then drastically cut memory and speed bottlenecks by successfully restricting the optimization to the last linear layer. This hinges on the unexpected finding that this layer alone reliably captures the second-order statistics of the per-sample gradients. The resulting surrogate problem integrates readily into standard optimizers like SGD and AdamW, and can be solved using a small number of GPU-friendly iterations. Crucially, the method exhibits favorable scaling properties, as the relative computational overhead shrinks as the model size or input grows. Experiments on image classification benchmarks confirm reduced per-sample interference and improved generalization.

Apostolos Avranas, Marios Kountouris

The conventional, widely used treatment of deep learning models as black boxes provides limited or no insights into the mechanisms that guide neural network decisions. Significant research effort has been dedicated to building interpretable models to address this issue. Most efforts either focus on the high-level features associated with the last layers, or attempt to interpret the output of a single layer. In this paper, we take a novel approach to enhance the transparency of the function of the whole network. We propose a neural network architecture for classification, in which the information that is relevant to each class flows through specific paths. These paths are designed in advance before training leveraging coding theory and without depending on the semantic similarities between classes. A key property is that each path can be used as an autonomous single-purpose model. This enables us to obtain, without any additional training and for any class, a lightweight binary classifier that has at least $60\%$ fewer parameters than the original network. Furthermore, our coding theory based approach allows the neural network to make early predictions at intermediate layers during inference, without requiring its full evaluation. Remarkably, the proposed architecture provides all the aforementioned properties while improving the overall accuracy. We demonstrate these properties on a slightly modified ResNeXt model tested on CIFAR-10/100 and ImageNet-1k.

Apostolos Avranas, Marios Kountouris, Philippe Ciblat

The problem of resource constrained scheduling in a dynamic and heterogeneous wireless setting is considered here. In our setup, the available limited bandwidth resources are allocated in order to serve randomly arriving service demands, which in turn belong to different classes in terms of payload data requirement, delay tolerance, and importance/priority. In addition to heterogeneous traffic, another major challenge stems from random service rates due to time-varying wireless communication channels. Various approaches for scheduling and resource allocation can be used, ranging from simple greedy heuristics and constrained optimization to combinatorics. Those methods are tailored to specific network or application configuration and are usually suboptimal. To this purpose, we resort to deep reinforcement learning (DRL) and propose a distributional Deep Deterministic Policy Gradient (DDPG) algorithm combined with Deep Sets to tackle the aforementioned problem. Furthermore, we present a novel way to use a Dueling Network, which leads to further performance improvement. Our proposed algorithm is tested on both synthetic and real data, showing consistent gains against state-of-the-art conventional methods from combinatorics, optimization, and scheduling metrics.

Anastasios Giovanidis, Apostolos Avranas

This article introduces a novel family of decentralised caching policies, applicable to wireless networks with finite storage at the edge-nodes (stations). These policies, that are based on the Least-Recently-Used replacement principle, are here referred to as spatial multi-LRU. They update cache inventories in a way that provides content diversity to users who are covered by, and thus have access to, more than one station. Two variations are proposed, the multi-LRU-One and -All, which differ in the number of replicas inserted in the involved caches. We analyse their performance under two types of traffic demand, the Independent Reference Model (IRM) and a model that exhibits temporal locality. For IRM, we propose a Che-like approximation to predict the hit probability, which gives very accurate results. Numerical evaluations show that the performance of multi-LRU increases the more the multi-coverage areas increase, and it is close to the performance of centralised policies, when multi-coverage is sufficient. For IRM traffic, multi-LRU-One is preferable to multi-LRU-All, whereas when the traffic exhibits temporal locality the -All variation can perform better. Both variations outperform the simple LRU. When popularity knowledge is not accurate, the new policies can perform better than centralised ones.