Antonios Valkanas

Large language models (LLMs) have achieved impressive results on complex reasoning tasks, but their high inference cost remains a major barrier to real-world deployment. A promising solution is to use cascaded inference, where small, cheap models handle easy queries, and only the hardest examples are escalated to more powerful models. However, existing cascade methods typically rely on supervised training with labeled data, offer no theoretical generalization guarantees, and provide limited control over test-time computational cost. We introduce C3PO (Cost Controlled Cascaded Prediction Optimization), a self-supervised framework for optimizing LLM cascades under probabilistic cost constraints. By focusing on minimizing regret with respect to the most powerful model (MPM), C3PO avoids the need for labeled data by constructing a cascade using only unlabeled model outputs. It leverages conformal prediction to bound the probability that inference cost exceeds a user-specified budget. We provide theoretical guarantees on both cost control and generalization error, and show that our optimization procedure is effective even with small calibration sets. Empirically, C3PO achieves state-of-the-art performance across a diverse set of reasoning benchmarks including GSM8K, MATH-500, BigBench-Hard and AIME, outperforming strong LLM cascading baselines in both accuracy and cost-efficiency. Our results demonstrate that principled, label-free cascade optimization can enable scalable LLM deployment.

Classifying nodes in a graph is a common problem. The ideal classifier must adapt to any imbalances in the class distribution. It must also use information in the clustering structure of real-world graphs. Existing Graph Neural Networks (GNNs) have not addressed both problems together. We propose the Enhanced Cluster-aware Graph Network (ECGN), a novel method that addresses these issues by integrating cluster-specific training with synthetic node generation. Unlike traditional GNNs that apply the same node update process for all nodes, ECGN learns different aggregations for different clusters. We also use the clusters to generate new minority-class nodes in a way that helps clarify the inter-class decision boundary. By combining cluster-aware embeddings with a global integration step, ECGN enhances the quality of the resulting node embeddings. Our method works with any underlying GNN and any cluster generation technique. Experimental results show that ECGN consistently outperforms its closest competitors by up to 11% on some widely studied benchmark datasets.

Adaptive importance sampling (AIS) methods provide a useful alternative to Markov Chain Monte Carlo (MCMC) algorithms for performing inference of intractable distributions. Population Monte Carlo (PMC) algorithms constitute a family of AIS approaches which adapt the proposal distributions iteratively to improve the approximation of the target distribution. Recent work in this area primarily focuses on ameliorating the proposal adaptation procedure for high-dimensional applications. However, most of the AIS algorithms use simple proposal distributions for sampling, which might be inadequate in exploring target distributions with intricate geometries. In this work, we construct expressive proposal distributions in the AIS framework using normalizing flow, an appealing approach for modeling complex distributions. We use an iterative parameter update rule to enhance the approximation of the target distribution. Numerical experiments show that in high-dimensional settings, the proposed algorithm offers significantly improved performance compared to the existing techniques.

Adaptive importance sampling (AIS) methods provide a useful alternative to Markov Chain Monte Carlo (MCMC) algorithms for performing inference of intractable distributions. Population Monte Carlo (PMC) algorithms constitute a family of AIS approaches which adapt the proposal distributions iteratively to improve the approximation of the target distribution. Recent work in this area primarily focuses on ameliorating the proposal adaptation procedure for high-dimensional applications. However, most of the AIS algorithms use simple proposal distributions for sampling, which might be inadequate in exploring target distributions with intricate geometries. In this work, we construct expressive proposal distributions in the AIS framework using normalizing flow, an appealing approach for modeling complex distributions. We use an iterative parameter update rule to enhance the approximation of the target distribution. Numerical experiments show that in high-dimensional settings, the proposed algorithm offers significantly improved performance compared to the existing techniques.

We show that the task of synthesizing human motion conditioned on a set of key frames can be solved more accurately and effectively if a deep learning based interpolator operates in the delta mode using the spherical linear interpolator as a baseline. We empirically demonstrate the strength of our approach on publicly available datasets achieving state-of-the-art performance. We further generalize these results by showing that the $\Delta$ -regime is viable with respect to the reference of the last known frame (also known as the zero-velocity model). This supports the more general conclusion that operating in the reference frame local to input frames is more accurate and robust than in the global (world) reference frame advocated in previous work. Our code is publicly available at: https://github.com/boreshkinai/delta-interpolator .

Recommender systems now consume large-scale data and play a significant role in improving user experience. Graph Neural Networks (GNNs) have emerged as one of the most effective recommender system models because they model the rich relational information. The ever-growing volume of data can make training GNNs prohibitively expensive. To address this, previous attempts propose to train the GNN models incrementally as new data blocks arrive. Feature and structure knowledge distillation techniques have been explored to allow the GNN model to train in a fast incremental fashion while alleviating the catastrophic forgetting problem. However, preserving the same amount of the historical information for all users is sub-optimal since it fails to take into account the dynamics of each user's change of preferences. For the users whose interests shift substantially, retaining too much of the old knowledge can overly constrain the model, preventing it from quickly adapting to the users’ novel interests. In contrast, for users who have static preferences, model performance can benefit greatly from preserving as much of the user's long-term preferences as possible. In this work, we propose a novel training strategy that adaptively learns personalized imitation weights for each user to balance the contribution from the recent data and the amount of knowledge to be distilled from previous time periods. We demonstrate the effectiveness of learning imitation weights via a comparison on five diverse datasets for three state-of-art structure distillation based recommender systems. The performance shows consistent improvement over competitive incremental learning techniques.

Recommender systems now consume large-scale data and play a significant role in improving user experience. Graph Neural Networks (GNNs) have emerged as one of the most effective recommender system models because they model the rich relational information. The ever-growing volume of data can make training GNNs prohibitively expensive. To address this, previous attempts propose to train the GNN models incrementally as new data blocks arrive. Feature and structure knowledge distillation techniques have been explored to allow the GNN model to train in a fast incremental fashion while alleviating the catastrophic forgetting problem. However, preserving the same amount of the historical information for all users is sub-optimal since it fails to take into account the dynamics of each user's change of preferences. For the users whose interests shift substantially, retaining too much of the old knowledge can overly constrain the model, preventing it from quickly adapting to the users' novel interests. In contrast, for users who have static preferences, model performance can benefit greatly from preserving as much of the user's long-term preferences as possible. In this work, we propose a novel training strategy that adaptively learns personalized imitation weights for each user to balance the contribution from the recent data and the amount of knowledge to be distilled from previous time periods. We demonstrate the effectiveness of learning imitation weights via a comparison on five diverse datasets for three state-of-art structure distillation based recommender systems. The performance shows consistent improvement over competitive incremental learning techninques.

We show that the task of synthesizing human motion conditioned on a set of key frames can be solved more accurately and effectively if a deep learning based interpolator operates in the delta mode using the spherical linear interpolator as a baseline. We empirically demonstrate the strength of our approach on publicly available datasets achieving state-of-the-art performance. We further generalize these results by showing that the $\Delta$-regime is viable with respect to the reference of the last known frame (also known as the zero-velocity model). This supports the more general conclusion that operating in the reference frame local to input frames is more accurate and robust than in the global (world) reference frame advocated in previous work. Our code is publicly available at https://github.com/boreshkinai/delta-interpolator.

Working with the abstract concepts of signals in time and frequency domains without a concrete tool to understand the very nature of these fundamental concepts has been a major impediment for undergraduate engineering students attending a course on Signals and Systems or Communications. Classroom teaching in such courses has been traditionally confined to working out the mathematics behind the concepts. Hence a tool is needed that allows students to explore the actual radio spectrum leading to a real appreciation of the subject matter. This paper considers the use of a low-cost RTL-SDR as an educational tool to address this difficulty. The approach presented can be used for in-class demonstrations and at-home activities aiding in a better understanding of the nature of signals in time and frequency domains. RTL-SDR can allow students to explore the radio spectrum on their own in an easy, cost-effective and academically enriching manner.