Dimitrios Stamoulis

Yanan Jian, Fuxun Yu, Simranjit Singh et al.

Aerial object detection is a challenging task, in which one major obstacle lies in the limitations of large-scale data collection and the long-tail distribution of certain classes. Synthetic data offers a promising solution, especially with recent advances in diffusion-based methods like stable diffusion (SD). However, the direct application of diffusion methods to aerial domains poses unique challenges: stable diffusion's optimization for rich ground-level semantics doesn't align with the sparse nature of aerial objects, and the extraction of post-synthesis object coordinates remains problematic. To address these challenges, we introduce a synthetic data augmentation framework tailored for aerial images. It encompasses sparse-to-dense region of interest (ROI) extraction to bridge the semantic gap, fine-tuning the diffusion model with low-rank adaptation (LORA) to circumvent exhaustive retraining, and finally, a Copy-Paste method to compose synthesized objects with backgrounds, providing a nuanced approach to aerial object detection through synthetic data.

Chaehong Lee, Varatheepan Paramanayakam, Andreas Karatzas et al.

We present GeoLLM-Squad, a geospatial Copilot that introduces the novel multi-agent paradigm to remote sensing (RS) workflows. Unlike existing single-agent approaches that rely on monolithic large language models (LLM), GeoLLM-Squad separates agentic orchestration from geospatial task-solving, by delegating RS tasks to specialized sub-agents. Built on the open-source AutoGen and GeoLLM-Engine frameworks, our work enables the modular integration of diverse applications, spanning urban monitoring, forestry protection, climate analysis, and agriculture studies. Our results demonstrate that while single-agent systems struggle to scale with increasing RS task complexity, GeoLLM-Squad maintains robust performance, achieving a 17% improvement in agentic correctness over state-of-the-art baselines. Our findings highlight the potential of multi-agent AI in advancing RS workflows.

Dimitrios Stamoulis, Ruizhou Ding, Di Wang et al.

Can we reduce the search cost of Neural Architecture Search (NAS) from days down to only few hours? NAS methods automate the design of Convolutional Networks (ConvNets) under hardware constraints and they have emerged as key components of AutoML frameworks. However, the NAS problem remains challenging due to the combinatorially large design space and the significant search time (at least 200 GPU-hours). In this work, we alleviate the NAS search cost down to less than 3 hours, while achieving state-of-the-art image classification results under mobile latency constraints. We propose a novel differentiable NAS formulation, namely Single-Path NAS, that uses one single-path over-parameterized ConvNet to encode all architectural decisions based on shared convolutional kernel parameters, hence drastically decreasing the search overhead. Single-Path NAS achieves state-of-the-art top-1 ImageNet accuracy (75.62%), hence outperforming existing mobile NAS methods in similar latency settings (~80ms). In particular, we enhance the accuracy-runtime trade-off in differentiable NAS by treating the Squeeze-and-Excitation path as a fully searchable operation with our novel single-path encoding. Our method has an overall cost of only 8 epochs (24 TPU-hours), which is up to 5,000x faster compared to prior work. Moreover, we study how different NAS formulation choices affect the performance of the designed ConvNets. Furthermore, we exploit the efficiency of our method to answer an interesting question: instead of empirically tuning the hyperparameters of the NAS solver (as in prior work), can we automatically find the hyperparameter values that yield the desired accuracy-runtime trade-off? We open-source our entire codebase at: https://github.com/dstamoulis/single-path-nas.

Dimitrios Stamoulis, Ruizhou Ding, Di Wang et al.

Can we automatically design a Convolutional Network (ConvNet) with the highest image classification accuracy under the runtime constraint of a mobile device? Neural architecture search (NAS) has revolutionized the design of hardware-efficient ConvNets by automating this process. However, the NAS problem remains challenging due to the combinatorially large design space, causing a significant searching time (at least 200 GPU-hours). To alleviate this complexity, we propose Single-Path NAS, a novel differentiable NAS method for designing hardware-efficient ConvNets in less than 4 hours. Our contributions are as follows: 1. Single-path search space: Compared to previous differentiable NAS methods, Single-Path NAS uses one single-path over-parameterized ConvNet to encode all architectural decisions with shared convolutional kernel parameters, hence drastically decreasing the number of trainable parameters and the search cost down to few epochs. 2. Hardware-efficient ImageNet classification: Single-Path NAS achieves 74.96% top-1 accuracy on ImageNet with 79ms latency on a Pixel 1 phone, which is state-of-the-art accuracy compared to NAS methods with similar constraints (<80ms). 3. NAS efficiency: Single-Path NAS search cost is only 8 epochs (30 TPU-hours), which is up to 5,000x faster compared to prior work. 4. Reproducibility: Unlike all recent mobile-efficient NAS methods which only release pretrained models, we open-source our entire codebase at: https://github.com/dstamoulis/single-path-nas.

Simranjit Singh, Michael Fore, Dimitrios Stamoulis

Geospatial Copilots unlock unprecedented potential for performing Earth Observation (EO) applications through natural language instructions. However, existing agents rely on overly simplified single tasks and template-based prompts, creating a disconnect with real-world scenarios. In this work, we present GeoLLM-Engine, an environment for tool-augmented agents with intricate tasks routinely executed by analysts on remote sensing platforms. We enrich our environment with geospatial API tools, dynamic maps/UIs, and external multimodal knowledge bases to properly gauge an agent's proficiency in interpreting realistic high-level natural language commands and its functional correctness in task completions. By alleviating overheads typically associated with human-in-the-loop benchmark curation, we harness our massively parallel engine across 100 GPT-4-Turbo nodes, scaling to over half a million diverse multi-tool tasks and across 1.1 million satellite images. By moving beyond traditional single-task image-caption paradigms, we investigate state-of-the-art agents and prompting techniques against long-horizon prompts.

Michael Fore, Simranjit Singh, Dimitrios Stamoulis

In this preliminary study, we investigate a GPT-driven intent-based reasoning approach to streamline tool selection for large language models (LLMs) aimed at system efficiency. By identifying the intent behind user prompts at runtime, we narrow down the API toolset required for task execution, reducing token consumption by up to 24.6\%. Early results on a real-world, massively parallel Copilot platform with over 100 GPT-4-Turbo nodes show cost reductions and potential towards improving LLM-based system efficiency.

Simranjit Singh, Andreas Karatzas, Michael Fore et al.

State-of-the-art sequential reasoning in Large Language Models (LLMs) has expanded the capabilities of Copilots beyond conversational tasks to complex function calling, managing thousands of API calls. However, the tendency of compositional prompting to segment tasks into multiple steps, each requiring a round-trip to the GPT APIs, leads to increased system latency and costs. Although recent advancements in parallel function calling have improved tool execution per API call, they may necessitate more detailed in-context instructions and task breakdown at the prompt level, resulting in higher engineering and production costs. Inspired by the hardware design principles of multiply-add (MAD) operations, which fuse multiple arithmetic operations into a single task from the compiler's perspective, we propose LLM-Tool Compiler, which selectively fuses similar types of tool operations under a single function at runtime, presenting them as a unified task to the LLM. This selective fusion inherently enhances parallelization and efficiency. Benchmarked on a large-scale Copilot platform, LLM-Tool Compiler achieves up to four times more parallel calls than existing methods, reducing token costs and latency by up to 40% and 12%, respectively.

Simranjit Singh, Michael Fore, Dimitrios Stamoulis

Tool-augmented Large Language Models (LLMs) have shown impressive capabilities in remote sensing (RS) applications. However, existing benchmarks assume question-answering input templates over predefined image-text data pairs. These standalone instructions neglect the intricacies of realistic user-grounded tasks. Consider a geospatial analyst: they zoom in a map area, they draw a region over which to collect satellite imagery, and they succinctly ask "Detect all objects here". Where is `here`, if it is not explicitly hardcoded in the image-text template, but instead is implied by the system state, e.g., the live map positioning? To bridge this gap, we present GeoLLM-QA, a benchmark designed to capture long sequences of verbal, visual, and click-based actions on a real UI platform. Through in-depth evaluation of state-of-the-art LLMs over a diverse set of 1,000 tasks, we offer insights towards stronger agents for RS applications.

Dimitrios Stamoulis, Diana Marculescu

Geospatial Copilots hold immense potential for automating Earth observation (EO) and climate monitoring workflows, yet their reliance on large-scale models such as GPT-4o introduces a paradox: tools intended for sustainability studies often incur unsustainable costs. Using agentic AI frameworks in geospatial applications can amass thousands of dollars in API charges or requires expensive, power-intensive GPUs for deployment, creating barriers for researchers, policymakers, and NGOs. Unfortunately, when geospatial Copilots are deployed with open language models (OLMs), performance often degrades due to their dependence on GPT-optimized logic. In this paper, we present Geo-OLM, a tool-augmented geospatial agent that leverages the novel paradigm of state-driven LLM reasoning to decouple task progression from tool calling. By alleviating the workflow reasoning burden, our approach enables low-resource OLMs to complete geospatial tasks more effectively. When downsizing to small models below 7B parameters, Geo-OLM outperforms the strongest prior geospatial baselines by 32.8% in successful query completion rates. Our method performs comparably to proprietary models achieving results within 10% of GPT-4o, while reducing inference costs by two orders of magnitude from \$500-\$1000 to under \$10. We present an in-depth analysis with geospatial downstream benchmarks, providing key insights to help practitioners effectively deploy OLMs for EO applications.

Andreas Karatzas, Dimitrios Stamoulis, Iraklis Anagnostopoulos

Modern edge data centers simultaneously handle multiple Deep Neural Networks (DNNs), leading to significant challenges in workload management. Thus, current management systems must leverage the architectural heterogeneity of new embedded systems to efficiently handle multi-DNN workloads. This paper introduces RankMap, a priority-aware manager specifically designed for multi-DNN tasks on heterogeneous embedded devices. RankMap addresses the extensive solution space of multi-DNN mapping through stochastic space exploration combined with a performance estimator. Experimental results show that RankMap achieves x3.6 higher average throughput compared to existing methods, while preventing DNN starvation under heavy workloads and improving the prioritization of specified DNNs by x57.5.

Panagiotis Michelakis, Yiannis Hadjiyiannis, Dimitrios Stamoulis

Evaluating AI agents that solve real-world tasks through function-call sequences remains an open challenge. Existing agentic benchmarks often reduce evaluation to a binary judgment of the final state, overlooking critical aspects such as safety, efficiency, and intermediate correctness. We propose a framework based on deterministic finite automata (DFAs) that encodes tasks as sets of valid tool-use paths, enabling principled assessment of agent behavior in diverse world models. Building on this foundation, we introduce CORE, a suite of five metrics, namely Path Correctness, Path Correctness - Kendall's tau Composite, Prefix Criticality, Harmful-Call Rate, and Efficiency, that quantify alignment with expected execution patterns. Across diverse worlds, our method reveals important performance differences between agents that would otherwise appear equivalent under traditional final-state evaluation schemes.

Amulya Bhattaram, Janani Ramamoorthy, Ranit Gupta et al.

The rise of agentic AI workflows unlocks novel opportunities for computer systems design and optimization. However, for specialized domains such as program synthesis, the relative scarcity of HDL and proprietary EDA resources online compared to more common programming tasks introduces challenges, often necessitating task-specific fine-tuning, high inference costs, and manually-crafted agent orchestration. In this work, we present VeriMaAS, a multi-agent framework designed to automatically compose agentic workflows for RTL code generation. Our key insight is to integrate formal verification feedback from HDL tools directly into workflow generation, reducing the cost of gradient-based updates or prolonged reasoning traces. Our method improves synthesis performance by 5-7% for pass@k over fine-tuned baselines, while requiring only a few hundred training examples, representing an order-of-magnitude reduction in supervision cost.

Amulya Bhattaram, Justin Chung, Stanley Chung et al.

We present GeoFlow, a method that automatically generates agentic workflows for geospatial tasks. Unlike prior work that focuses on reasoning decomposition and leaves API selection implicit, our method provides each agent with detailed tool-calling objectives to guide geospatial API invocation at runtime. GeoFlow increases agentic success by 6.8% and reduces token usage by up to fourfold across major LLM families compared to state-of-the-art approaches.

Varatheepan Paramanayakam, Andreas Karatzas, Iraklis Anagnostopoulos et al.

Large Language Models (LLMs) enable real-time function calling in edge AI systems but introduce significant computational overhead, leading to high power consumption and carbon emissions. Existing methods optimize for performance while neglecting sustainability, making them inefficient for energy-constrained environments. We introduce CarbonCall, a sustainability-aware function-calling framework that integrates dynamic tool selection, carbon-aware execution, and quantized LLM adaptation. CarbonCall adjusts power thresholds based on real-time carbon intensity forecasts and switches between model variants to sustain high tokens-per-second throughput under power constraints. Experiments on an NVIDIA Jetson AGX Orin show that CarbonCall reduces carbon emissions by up to 52%, power consumption by 30%, and execution time by 30%, while maintaining high efficiency.

Varatheepan Paramanayakam, Andreas Karatzas, Dimitrios Stamoulis et al.

Edge computing systems struggle to efficiently manage multiple concurrent deep neural network (DNN) workloads while meeting strict latency requirements, minimizing power consumption, and maintaining environmental sustainability. This paper introduces Ecomap, a sustainability-driven framework that dynamically adjusts the maximum power threshold of edge devices based on real-time carbon intensity. Ecomap incorporates the innovative use of mixed-quality models, allowing it to dynamically replace computationally heavy DNNs with lighter alternatives when latency constraints are violated, ensuring service responsiveness with minimal accuracy loss. Additionally, it employs a transformer-based estimator to guide efficient workload mappings. Experimental results using NVIDIA Jetson AGX Xavier demonstrate that Ecomap reduces carbon emissions by an average of 30% and achieves a 25% lower carbon delay product (CDP) compared to state-of-the-art methods, while maintaining comparable or better latency and power efficiency.

Emmanuel O. Badmus, Peng Sang, Dimitrios Stamoulis et al.

Rapid electrification and decarbonization are increasing the complexity of distribution grid (DG) operation and planning, necessitating advanced computational analyses to ensure reliability and resilience. These analyses depend on disparate workflows comprising complex models, function calls, and data pipelines that require substantial expert knowledge and remain difficult to automate. Workforce and budget constraints further limit utilities' ability to apply such analyses at scale. To address this gap, we build an agentic system PowerChain, which is capable of autonomously performing complex grid analyses. Existing agentic AI systems are typically developed in a bottom-up manner with customized context for predefined analysis tasks; therefore, they do not generalize to tasks that the agent has never seen. In comparison, to generalize to unseen DG analysis tasks, PowerChain dynamically generates structured context by leveraging supervisory signals from self-contained power systems tools (e.g., GridLAB-D) and an optimized set of expert-annotated and verified reasoning trajectories. For complex DG tasks defined in natural language, empirical results on real utility data demonstrate that PowerChain achieves up to a 144/% improvement in performance over baselines.

Varatheepan Paramanayakam, Andreas Karatzas, Iraklis Anagnostopoulos et al.

The advanced function-calling capabilities of foundation models open up new possibilities for deploying agents to perform complex API tasks. However, managing large amounts of data and interacting with numerous APIs makes function calling hardware-intensive and costly, especially on edge devices. Current Large Language Models (LLMs) struggle with function calling at the edge because they cannot handle complex inputs or manage multiple tools effectively. This results in low task-completion accuracy, increased delays, and higher power consumption. In this work, we introduce Less-is-More, a novel fine-tuning-free function-calling scheme for dynamic tool selection. Our approach is based on the key insight that selectively reducing the number of tools available to LLMs significantly improves their function-calling performance, execution time, and power efficiency on edge devices. Experimental results with state-of-the-art LLMs on edge hardware show agentic success rate improvements, with execution time reduced by up to 70% and power consumption by up to 40%.

Simranjit Singh, Michael Fore, Andreas Karatzas et al.

As Large Language Models (LLMs) broaden their capabilities to manage thousands of API calls, they are confronted with complex data operations across vast datasets with significant overhead to the underlying system. In this work, we introduce LLM-dCache to optimize data accesses by treating cache operations as callable API functions exposed to the tool-augmented agent. We grant LLMs the autonomy to manage cache decisions via prompting, seamlessly integrating with existing function-calling mechanisms. Tested on an industry-scale massively parallel platform that spans hundreds of GPT endpoints and terabytes of imagery, our method improves Copilot times by an average of 1.24x across various LLMs and prompting techniques.

Fuxun Yu, Dimitrios Stamoulis, Di Wang et al.

This paper gives an overview of our ongoing work on the design space exploration of efficient deep neural networks (DNNs). Specifically, we cover two aspects: (1) static architecture design efficiency and (2) dynamic model execution efficiency. For static architecture design, different from existing end-to-end hardware modeling assumptions, we conduct full-stack profiling at the GPU core level to identify better accuracy-latency trade-offs for DNN designs. For dynamic model execution, different from prior work that tackles model redundancy at the DNN-channels level, we explore a new dimension of DNN feature map redundancy to be dynamically traversed at runtime. Last, we highlight several open questions that are poised to draw research attention in the next few years.

Dimitrios Stamoulis, Ruizhou Ding, Di Wang et al.

Can we automatically design a Convolutional Network (ConvNet) with the highest image classification accuracy under the latency constraint of a mobile device? Neural Architecture Search (NAS) for ConvNet design is a challenging problem due to the combinatorially large design space and search time (at least 200 GPU-hours). To alleviate this complexity, we propose Single-Path NAS, a novel differentiable NAS method for designing device-efficient ConvNets in less than 4 hours. 1. Novel NAS formulation: our method introduces a single-path, over-parameterized ConvNet to encode all architectural decisions with shared convolutional kernel parameters. 2. NAS efficiency: Our method decreases the NAS search cost down to 8 epochs (30 TPU-hours), i.e., up to 5,000x faster compared to prior work. 3. On-device image classification: Single-Path NAS achieves 74.96% top-1 accuracy on ImageNet with 79ms inference latency on a Pixel 1 phone, which is state-of-the-art accuracy compared to NAS methods with similar latency (<80ms).

Diana Marculescu, Dimitrios Stamoulis, Ermao Cai

Recent breakthroughs in Deep Learning (DL) applications have made DL models a key component in almost every modern computing system. The increased popularity of DL applications deployed on a wide-spectrum of platforms have resulted in a plethora of design challenges related to the constraints introduced by the hardware itself. What is the latency or energy cost for an inference made by a Deep Neural Network (DNN)? Is it possible to predict this latency or energy consumption before a model is trained? If yes, how can machine learners take advantage of these models to design the hardware-optimal DNN for deployment? From lengthening battery life of mobile devices to reducing the runtime requirements of DL models executing in the cloud, the answers to these questions have drawn significant attention. One cannot optimize what isn't properly modeled. Therefore, it is important to understand the hardware efficiency of DL models during serving for making an inference, before even training the model. This key observation has motivated the use of predictive models to capture the hardware performance or energy efficiency of DL applications. Furthermore, DL practitioners are challenged with the task of designing the DNN model, i.e., of tuning the hyper-parameters of the DNN architecture, while optimizing for both accuracy of the DL model and its hardware efficiency. Therefore, state-of-the-art methodologies have proposed hardware-aware hyper-parameter optimization techniques. In this paper, we provide a comprehensive assessment of state-of-the-art work and selected results on the hardware-aware modeling and optimization for DL applications. We also highlight several open questions that are poised to give rise to novel hardware-aware designs in the next few years, as DL applications continue to significantly impact associated hardware systems and platforms.

Dimitrios Stamoulis, Ting-Wu Chin, Anand Krishnan Prakash et al.

As convolutional neural networks (CNNs) enable state-of-the-art computer vision applications, their high energy consumption has emerged as a key impediment to their deployment on embedded and mobile devices. Towards efficient image classification under hardware constraints, prior work has proposed adaptive CNNs, i.e., systems of networks with different accuracy and computation characteristics, where a selection scheme adaptively selects the network to be evaluated for each input image. While previous efforts have investigated different network selection schemes, we find that they do not necessarily result in energy savings when deployed on mobile systems. The key limitation of existing methods is that they learn only how data should be processed among the CNNs and not the network architectures, with each network being treated as a blackbox. To address this limitation, we pursue a more powerful design paradigm where the architecture settings of the CNNs are treated as hyper-parameters to be globally optimized. We cast the design of adaptive CNNs as a hyper-parameter optimization problem with respect to energy, accuracy, and communication constraints imposed by the mobile device. To efficiently solve this problem, we adapt Bayesian optimization to the properties of the design space, reaching near-optimal configurations in few tens of function evaluations. Our method reduces the energy consumed for image classification on a mobile device by up to 6x, compared to the best previously published work that uses CNNs as blackboxes. Finally, we evaluate two image classification practices, i.e., classifying all images locally versus over the cloud under energy and communication constraints.

Dimitrios Stamoulis, Ermao Cai, Da-Cheng Juan et al.

While selecting the hyper-parameters of Neural Networks (NNs) has been so far treated as an art, the emergence of more complex, deeper architectures poses increasingly more challenges to designers and Machine Learning (ML) practitioners, especially when power and memory constraints need to be considered. In this work, we propose HyperPower, a framework that enables efficient Bayesian optimization and random search in the context of power- and memory-constrained hyper-parameter optimization for NNs running on a given hardware platform. HyperPower is the first work (i) to show that power consumption can be used as a low-cost, a priori known constraint, and (ii) to propose predictive models for the power and memory of NNs executing on GPUs. Thanks to HyperPower, the number of function evaluations and the best test error achieved by a constraint-unaware method are reached up to 112.99x and 30.12x faster, respectively, while never considering invalid configurations. HyperPower significantly speeds up the hyper-parameter optimization, achieving up to 57.20x more function evaluations compared to constraint-unaware methods for a given time interval, effectively yielding significant accuracy improvements by up to 67.6%.

Ermao Cai, Da-Cheng Juan, Dimitrios Stamoulis et al.

"How much energy is consumed for an inference made by a convolutional neural network (CNN)?" With the increased popularity of CNNs deployed on the wide-spectrum of platforms (from mobile devices to workstations), the answer to this question has drawn significant attention. From lengthening battery life of mobile devices to reducing the energy bill of a datacenter, it is important to understand the energy efficiency of CNNs during serving for making an inference, before actually training the model. In this work, we propose NeuralPower: a layer-wise predictive framework based on sparse polynomial regression, for predicting the serving energy consumption of a CNN deployed on any GPU platform. Given the architecture of a CNN, NeuralPower provides an accurate prediction and breakdown for power and runtime across all layers in the whole network, helping machine learners quickly identify the power, runtime, or energy bottlenecks. We also propose the "energy-precision ratio" (EPR) metric to guide machine learners in selecting an energy-efficient CNN architecture that better trades off the energy consumption and prediction accuracy. The experimental results show that the prediction accuracy of the proposed NeuralPower outperforms the best published model to date, yielding an improvement in accuracy of up to 68.5%. We also assess the accuracy of predictions at the network level, by predicting the runtime, power, and energy of state-of-the-art CNN architectures, achieving an average accuracy of 88.24% in runtime, 88.34% in power, and 97.21% in energy. We comprehensively corroborate the effectiveness of NeuralPower as a powerful framework for machine learners by testing it on different GPU platforms and Deep Learning software tools.