Entries by NEC Labs America

DWIM: Towards Tool-aware Visual Reasoning via Discrepancy-aware Workflow Generation & Instruct-Masking Tuning

/in /by

Visual reasoning (VR), which is crucial in many fields for enabling human-like visual understanding, remains highly challenging. Recently, compositional visual reasoning approaches, which leverage the reasoning abilities of large language models (LLMs) with integrated tools to solve problems, have shown promise as more effective strategies than end-to-end VR methods. However, these approaches face limitations, as frozen LLMs lack tool awareness in VR, leading to performance bottlenecks. While leveraging LLMs for reasoning is widely used in other domains, they are not directly applicable to VR due to limited training data, imperfect tools that introduce errors and reduce data collection efficiency in VR, and challenging in fine-tuning on noisy workflows. To address these challenges, we propose DWIM: i) Discrepancy-aware training Workflow generation, which assesses tool usage and extracts more viable workflows for training; and ii) Instruct-Masking fine-tuning, which guides the model to only clone effective actions, enabling the generation of more practical solutions. Our experiments demonstrate that DWIM achieves state-of-the-art performance across various VR tasks, exhibiting strong generalization on multiple widely-used datasets.

NEC Labs America Attends OFC 2025 in San Francisco

/in /by

The NEC Labs America Optical Networking and Sensing team is attending the 2025 Optical Fiber Communications Conference and Exhibition (OFC), the premier global event for optical networking and communications. Bringing together over 13,500 attendees from 83+ countries, more than 670 exhibitors, and hundreds of sessions featuring industry leaders, OFC 2025 serves as the central hub for innovation and collaboration in the field. At this year’s conference, NEC Labs America will showcase its cutting-edge research and advancements through multiple presentations, demonstrations, and workshops.

Free-Space Optical Sensing Using Vector Beam Spectra

/in /by

Vector beams are spatial modes that have spatially inhomogeneous states of polarization. Any light beam is a linear combination of vector beams, the coefficients of which comprise a vector beam “spectrum.” In this work, through numerical calculations, a novel method of free-space optical sensing is demonstrated using vector beam spectra, which are shown to be experimentally measurable via Stokes polarimetry. As proof of concept, vector beam spectra are numerically calculated for various beams and beam obstructions.

400-Gb/s mode division multiplexing-based bidirectional free space optical communication in real-time with commercial transponders

/in /by

In this work, for the first time, we experimentally demonstrate mode division multiplexing-based bidirectional free space optical communication in real-time using commercial transponders. As proof of concept, via bidirectional pairs of Hermite-Gaussian modes (HG00, HG10, and HG01), using a Telecom Infra Project Phoenix compliant commercial 400G transponder, 400-Gb/s data signals (56-Gbaud, DP-16QAM) are bidirectionally transmitted error free, i.e., with less than 1e-2 pre-FEC BERs, over approximately 1-m of free space

EdgeSync: Efficient Edge-Assisted Video Analytics via Network Contention-Aware Scheduling

/in /by

With the advancement of 5G, edge-assisted video analytics has become increasingly popular, driven by the technology’s ability to support low-latency, high-bandwidth applications. However, in scenarios where multiple clients competing for network resources, network contention poses a significant challenge. In this paper, we propose a novel scheduling algorithm that intelligently batches and aligns the offloading of multiple video analytics clients to optimize both network and edge server resource utilization while meeting the Service Level Objective (SLO). Experiment with a cellular network testbed shows that our approach successfully processes 93% or more of inference requests from 7 different clients to the edge server while meeting the SLOs, whereas other approaches achieve a lower success rate, ranging from 65% to 85% under the same condition.

Attribute-Centric Compositional Text-to-Image Generation

/in /by

Despite the recent impressive breakthroughs in text-to-image generation, generative models have difficulty in capturing thedata distribution of underrepresented attribute compositions while over-memorizing overrepresented attribute compositions,which raises public concerns about their robustness and fairness. To tackle this challenge, we propose ACTIG, an attributecentriccompositional text-to-image generation framework. We present an attribute-centric feature augmentation and a novelimage-free training scheme, which greatly improves model’s ability to generate images with underrepresented attributes.Wefurther propose an attribute-centric contrastive loss to avoid overfitting to overrepresented attribute compositions.We validateour framework on the CelebA-HQ and CUB datasets. Extensive experiments show that the compositional generalization ofACTIG is outstanding, and our framework outperforms previous works in terms of image quality and text-image consistency

G-Litter Marine Litter Dataset Augmentation with Diffusion Models and Large Language Models on GPU Acceleration

/in /by

Marine litter detection is crucial for environmental monitoring, yet the imbalance in existing datasets limits model performance in identifying various types of waste accurately. This paper presents an efficient data augmentation pipeline that combines generative diffusion models (e.g., Stable Diffusion) and Large Language Models (LLMs) to expand the G-Litter dataset, a marine litter dataset designed for autonomous detection in heterogeneous environments. Leveraging scalable diffusion models for image generation and Alpaca LLMs for diverse prompt generation, our approach augments underrepresented classes by generating over 200 additional images per class, significantly improving the dataset’s balance. Training G-Litter augmented dataset using YOLOv8 for object detection demonstrated an increase in detection performance, improving recall by 7.82% and mAP50 by 3.87% (compared with baseline results). This study emphasizes the potential for combining generative AI with HPC resources to automate data augmentation on large-scale, unstructured datasets, particularly in edge computing contexts for real-time marine monitoring. The models were tested on real videos captured during simulated missions, demonstrating a superior ability to detect submerged objects in dynamic scenarios. These results highlight the potential of generative AI techniques to improve dataset quality and detection model performance, laying the foundation for further expansion in real-time marine monitoring.

Exploiting VLM Localizability and Semantics for Open Vocabulary Action Detection

/in /by

Action detection aims to detect (recognize and localize) human actions spatially and temporally in videos. Existing approaches focus on the closed-set setting where an action detector is trained and tested on videos from a fixed set of action categories. However, this constrained setting is not viable in an open world where test videos inevitably come beyond the trained action categories. In this paper, we address the practical yet challenging Open-Vocabulary Action Detection (OVAD) problem. It aims to detect any action in test videos while training a model on a fixed set of action categories. To achieve such an open-vocabulary capability, we propose a novel method OpenMixer that exploits the inherent semantics and localizability of large vision-language models (VLM) within the family of query-based detection transformers (DETR). Specifically, the OpenMixer is developed by spatial and temporal OpenMixer blocks (S-OMBand T-OMB), and a dynamically fused alignment (DFA) module. The three components collectively enjoy the merits of strong generalization from pre-trained VLMs and end to-end learning from DETR design. Moreover, we established OVAD benchmarks under various settings, and the experimental results show that the OpenMixer performs the best over baselines for detecting seen and unseen actions.

Learning Disentangled Equivariant Representation for Explicitly Controllable 3D Molecule Generation

/in /by

We consider the conditional generation of 3D drug-like molecules with explicit control over molecular properties such as drug-like properties (e.g., Quantitative Estimate of Druglikenessor Synthetic Accessibility score) and effectively binding to specific protein sites. To tackle this problem, we propose an E(3)-equivariant Wasserstein autoencoder and factorize thelatent space of our generative model into two disentangled aspects: molecular properties and the remaining structural context of 3D molecules. Our model ensures explicit control over these molecular attributes while maintaining equivariance of coordinate representation and invariance of data likelihood. Furthermore, we introduce a novel alignment-based coordinate loss to adapt equivariant networks for auto-regressive denovo 3D molecule generation from scratch. Extensive experiments validate our model’s effectiveness on property-guidedand context-guided molecule generation, both for de-novo 3D molecule design and structure-based drug discovery against protein targets.

TimeCAP: Learning to Contextualize, Augment, and Predict Time Series Events with Large Language Model Agents

/in /by

Time series data is essential in various applications, including climate modeling, healthcare monitoring, and financial analytics. Understanding the contextual information associated with real-world time series data is often essential for accurate and reliable event predictions. In this paper, we introduce TimeCAP, a time-series processing framework that creatively employs Large Language Models (LLMs) as contextualizers of time series data, extending their typical usage as predictors. TimeCAP incorporates two independent LLM agents: one generates a textual summary capturing the context of the time series, while the other uses this enriched summary to make more informed predictions. In addition, TimeCAP employs a multi-modal encoder that synergizes with the LLM agents, enhancing predictive performance through mutual augmentation of inputs with in-context examples. Experimental results on real-world datasets demonstrate that TimeCAP outperforms state-of-the-art methods for time series event prediction, including those utilizing LLMs as predictors, achieving an average improvement of 28.75% in F1 score.