Abstract

Deep learning models are increasingly being used at particle colliders for various tasks like jet classification, anomaly detection, detector unfolding, etc. In this thesis, I investigate the efficacy of three machine learning models using various different representations of collider data as per the models’ strengths. I compare boosted decision trees, deep neural networks, and graph neural networks for the event classification task of identifying a typical Beyond Standard Model (BSM) signal coming from 𝑇𝑒𝑉-scale particles. The signal of interest to us here is that coming from a pair-produced heavy bottom quark partner, denoted by 𝐵, which decays through a new pseudoscalar/ scalar boson, denoted 𝜙, similar to the Higgs boson in the Standard Model. Such signatures have been gaining interest recently; Vector Like Quarks (VLQs) themselves are being searched for by the prominent collider experiments ATLAS and CMS at the LHC, but their signatures elude us so far. We perform the analyses for a significantly challenging final state where the subsequent decays of BSM particles are fully hadronic, that is, consisting of only objects arising from quark and gluon jet hadronization; no objects like electrons, photons, or muons exist to give a good handle on the signal. We see that basic graph neural networks significantly outperform other classical ML models. We further improve our results by using methods that can only be deployed in a graph neural network. Therefore, we conclude we can enhance future collider searches for 𝑇𝑒𝑉-scale particle arising from new physics phenomena. The research methodology leverages state-of-the-art simulation tools, including MadGraph5, Pythia8, and Delphes3. We scan over a wide range of parameter configuration to perform robust evaluations of the ML/DL models and propose novel training strategies, opening new avenues for sophisticated particle detection techniques in high-energy collider experiments.

Abstract

Spoken grammar refers to the grammatical structure in the spoken language. It is essential for clear and effective oral communication. Thus, it is an important skill in language acquisition. Spoken Grammar Error Detection (SGED) involves identifying grammatical errors in spoken utterances, and is a necessary component of computer-assisted language learning (CALL) systems that aim to provide learners with automatic feedback. Traditionally, SGED has been implemented through a cascaded pipeline where Automatic Speech Recognition (ASR) converts speech to text, followed by a text-based grammatical error detector (GED). However, this approach is limited by ASR error propagation, where recognition errors and language-model biases can overwrite or mask the learner’s actual grammatical mistakes, leading to sub-optimal performance. In this work, we systematically study statistical and end-to-end ASR systems for SGED under three speaking practices, which are commonly employed in CALL systems: memorised, semi-spontaneous, and spontaneous speaking practices. In memorised and semi-spontaneous speaking conditions, the intended grammatically correct text is known in advance, allowing us to study how different ASR-derived cues behave when learners produce incorrect forms. In the statistical ASR setting, we analyse pronunciation and alignment-based features such as Goodness of Pronunciation (GoP) scores and Force Alignment (FA) likelihoods, using the known grammatically correct text reference to examine how these cues vary with grammatical deviations. In the end-to-end ASR case, we similarly use Connectionist Temporal Classification (CTC) aligner probabilities computed against the ground truth grammatically correct text to understand how alignment behaviour changes when the spoken utterance deviates from the expected grammatical form. For spontaneous speaking condition, where no reference text is available, most of the existing works rely on the traditional cascaded system consisting of ASR and GED approach relying on the ASR decoded text; however, this design is constrained by ASR error propagation and language-model bias. To overcome the inherent limitations of cascaded SGED, we propose a method that directly transforms raw speech into a textual embedding space, termed unified embeddings, for SGED in two settings: (1) an end-to-end approach that predicts grammaticality directly from speech, and (2) an end-to-end fusionbased approach that combines unified embeddings with ASR transcript embeddings to leverage both modalities jointly. The unified embeddings are obtained by distilling knowledge from a pre-trained text encoder into a speech encoder via speech-text unification. To the best of our knowledge, this is the first work to explore both end-to-end and fusion-based approaches for the SGED task. Experiments show that the proposed unification-based fusion approach outperforms the cascaded baseline and endto-end SGED, highlighting that unified embeddings act as an effective complementary information for mitigating ASR errors in SGED

Abstract

As urbanization accelerates, the need for energy-efficient, reliable, and scalable smart city infrastructure has become increasingly critical. Street lighting systems form a ubiquitous urban asset and present an opportunity to serve not only as illumination infrastructure but also as a communication backbone for large-scale Internet of Things (IoT) deployments. However, existing smart street lighting solutions often suffer from limited scalability, lack of interoperability, high operational costs, or poor reliability in dense urban and non-line-of-sight environments like industrial areas. This thesis presents the design, implementation, and real-world deployment of a Wi-SUN (Wireless Smart Ubiquitous Network)–based smart street lighting system integrated with the oneM2M middleware platform. Wi-SUN, based on the IEEE 802.15.4g standard, operates in the license-free sub-GHz band and employs a self-forming, self-healing mesh topology using RPL-based Destination Oriented Directed Acyclic Graphs (DODAGs), making it well suited for large-scale urban IoT networks. The proposed system provides IP-addressable street lighting nodes capable of reliable multi-hop communication, adaptive control, and seamless cloud integration. A complete end-to-end system was designed, including custom edge hardware, firmware for Wi-SUN nodes, a border router, and backend integration using oneM2M. The system was deployed as a real-world testbed consisting of 30 smart streetlight nodes across the IIIT Hyderabad campus. Comprehensive experiments were conducted to evaluate network performance, including signal propagation and path loss analysis in indoor and outdoor urban environments, latency, throughput, DODAG formation time, and DODAG healing time under node failure scenarios. Experimental results demonstrate that Wi-SUN provides robust and reliable connectivity in urban non-line-of-sight conditions, with effective mesh self-healing and acceptable network convergence times for practical deployments. The integration with oneM2M enables interoperability with other smart city verticals and supports future extensibility. This work establishes a scalable, low-cost,and resilient smart street lighting architecture and serves as a practical testbed for future smart city IoT applications.

Abstract

The continued scaling of CMOS and FinFET technologies has introduced significant challenges in digital circuit design, including increased susceptibility to aging effects and heightened leakage power variability. This thesis presents a comprehensive exploration of machine learning (ML)-integrated methodologies for standard cell characterization and complex circuit analysis in the context of advanced CMOS and FinFET technologies. Through a sequence of these studies, the work addresses key challenges associated with reliability, power efficiency, and aging-aware modeling in nanoscale digital circuits. RelOps demonstrates an ML-assisted, multi-objective optimization framework for 16nm HP FinFET standard cells. By integrating SPICE-driven time-series datasets with accurate regression models, the framework enables efficient tuning of critical FinFET dimensions (e.g., channel length, fin width, and fin height), thereby mitigating aging effects such as NBTI and HCI across PVT corners. This approach achieves substantial improvements in power-delay product (PDP) and reduces simulation overhead, establishing a scalable path for performance-centric design optimization. PIDArc proposes a novel physics-informed machine learning (PIML) architecture based on a deaggregated methodology to model the aging-induced leakage behavior in 7nm and 10nm FinFETs through pattern knowledge. This approach addresses the limitations of conventional ML models in handling high-variance leakage data and captures the complex non-linear relationship between aging, PVT variations, and leakage power. The proposed method significantly outperforms existing deep learning and boosting models in terms of prediction accuracy, offering a robust solution for aging-aware leakage estimation. Optimo proposes an ML-based optimization at the complex circuit level via the OptiMo methodology, which automates the reduction of power and delay across diverse designs in CMOS technology. The experimental evaluations on multi-stage circuits demonstrate up to 64.6% PDP reduction, reinforcing the scalability and adaptability of the ML optimization framework from standard cells to hierarchical digital systems. Collectively, the proposed techniques establish a unified ML-driven design flow that bridges devicelevel physics, circuit-level modeling, and system-level optimization. This work not only advances the state-of-the-art in FinFET-based digital circuit design but also offers practical and computationally efficient tools for reliability-aware characterization and performance enhancement in emerging technology nodes.

Abstract

With the rise of quantum computing, the traditional cryptographic techniques may face potential vulnerabilities, particularly in the security-critical applications, such as Internet of Things (IoT)-enabled blockchain systems. These systems rely heavily on the digital signatures for authentication, data integrity, and secure transactions. However, the quantum algorithms like Shor’s algorithm threaten conventional public-key based cryptographic schemes, such as RSA and elliptic curve cryptography (ECC) which make post-quantum cryptography (PQC) essential for future-proof security. IoT devices often have constrained computational power and storage, and thus, it makes challenging to integrate complex security protocols. Post-quantum signature schemes need to be designed which are resilient against quantum attacks, such as lattice reduction attack, hybrid classical-quantum attacks, side-channel and fault injection attacks, and brute-force attacks (e.g., searching for secret keys) using the quantum Grover’s Search algorithm in the context of lattice-based signature schemes. Thus, these signature schemes should ensure the long-term security of IoT networks without significantly increasing computational overhead. Additionally, blockchain-based applications rely on immutable records, which become vulnerable if past transactions can be decrypted in the future. Moreover, post-quantum digital signatures mitigate this risk by providing quantum-safe authentication and verification, maintaining trust in blockchain networks. In addition, regulatory and compliance frameworks are evolving to address quantum security threats, necessitating the adoption of PQC in IoT-blockchain systems. Organizations deploying IoT solutions in critical sectors, such as healthcare, finance, and smart cities, must adopt quantum-resistant cryptographic methods to ensure data privacy and system integrity. This thesis aims to design and develop blockchain-integrated lattice-based cryptographic schemes, including aggregate signatures, multi-signatures, and Ciphertext-Policy AttributeBased Encryption (CP-ABE) in IoT environments, to improve security, efficiency, and scalability. These schemes are intended to address the specific challenges of IoT networks, where devices are often limited in resources and require lightweight but strong security measures. In the first contribution, we designed a new lattice-based aggregate signature scheme that relies on the complexity of the Ring Learning-with-Errors (Ring-LWE) lattice-based hard problem for security. This scheme is then applied within the Internet of Drones (IoD) systems using the blockchain technology to ensure secure and transparent data storage. In this scheme, we described the comprehensive security analysis and comparative study, which indicates that the proposed scheme offers enhanced security, including quantum resistance, is efficient. We implemented the testbed experiments and blockchain simulations further to confirm that theproposed scheme is viable for real-world drone applications. Next, by utilizing the widely-adopted lattice hardness assumptions, namely Ring-based short integer solutions (Ring-SIS) and Ring-LWE, we designed an efficient multi-signature scheme for blockchain-based IoT applications. Additionally, we incorporated the single round online phase in the signing algorithm to achieve a low round complexity. The blockchain has been used as an add-on service for secure data storage purposes. In this case, each signer engages in pre-processing the offline phase before the single round online phase. In the offline phase, all the signers engage in interactions and exchange a set of commit messages among themselves. Subsequently, each party employs a random linear combination approach to aggregate all the commit messages and utilize them to generate the corresponding signature of the signer. Without the participation of all signers, a multi-signature cannot be generated, thereby making it impossible for an adversary to produce a fraudulent multi-signature without knowledge of all signers’ secret keys. The security analysis confirms that the scheme is capable of withstanding various attacks, including quantum attacks. A blockchain simulation was conducted to measure the computational time required for mining blocks in the blockchain. Moreover, a comparative study was presented, featuring several existing lattice-based multisignature schemes, to showcase the effectiveness of the proposed scheme in IoT applications. Finally, we designed a novel multi-authority CP-ABE scheme based on the lattice structure in a distributed environment for IoT-based smart healthcare applications. The work supports the implementation of the trapdoor generation to substantially reduce the computational overhead associated with the keygen phase and Gaussian Pre-image sampling techniques. The proposed scheme that has been developed regards the numerous auth

Abstract

In decision-making settings such as medical diagnosis, underwriting, or sentencing in a court of law, decisions are often influenced by multiple factors like the individual’s background, experience, and personal biases. Variability in decisions across individuals is thus inevitable, and so are disparities in different systems. Every system has its own way of addressing prevailing disparities—be it through noise audits, standardized guidelines, or other approaches. In the medical field, for instance, prior studies have documented significant inter-observer variations in the interpretation of clinical images such as MRIs, X-rays, etc., leading to inconsistencies in diagnoses and treatments. To address the issue, the medical domain is witnessing active development of AI tools intended to support clinical data analysis and reduce disparities in healthcare outcomes. In this thesis, we explore a data cube-based methodology to explore the issue of disparities in the legal domain. In the legal domain, decisions related to parole or bail grants, child custody, or sentence imposition are often left to the judges’ discretion. Specifically for sentence imposition, guidelines in many countries around the world allow for subjectivity. For example, in India, while sentencing guidelines prescribe minimum and maximum punishments for different offences, the weights assigned to various aggravating and mitigating circumstances are left to the judge’s discretion. This flexibility, though intended to accommodate case-specific nuances, increases variance, leading to inconsistencies in trial outcomes, sentence lengths, and penalties across courts and judges. The literature widely acknowledges that anomalies and disparities exist in sentencing and other legal decisions, often stemming from personal beliefs, biases, and contextual factors. Over the years, several efforts have also been made to analyse and assess sentencing anomalies/disparities in India and other parts of the world, particularly with respect to individual factors such as gender, race, socioeconomic background, etc., through surveys, case studies, and machine learning techniques. Notably, the Online analytical processing (OLAP) methodology has been widely employed in literature to analyse multidimensional data in different domains. The concept of a data cube was proposed to summarise and extract all subcubes from a table, enabling multi-dimensional analysis and derivation of insights from diverse perspectives. It is commonly adopted to extract interesting trends and anomalies from multidimensional data in domains like sales, marketing, etc. However, so far, no effort has been made to extend the data cube-based framework to explore anomalies and disparities in the legal domain. In this thesis, we leverage the OLAP framework and propose a data cube-based approach to explore potential trends and anomalies in judicial decisions, particularly sentences. A major bottleneck in this domain has been the lack of structured datasets. To address this, we employed a large language model (LLM) to curate a structured dataset from unstructured data extracted manually from Indian criminal case judgments. We designed a conceptual schema by identifying relevant attributes, hierarchies, and defining appropriate aggregate measures. We used this schema to build a data cube on the curated dataset and facilitate anomaly detection. This approach enabled us to uncover potential anomalies in court sentences, particularly in terms of quantum and monetary penalties for similar offenses across Indian states. For instance, we observed that Kerala imposed relatively higher monetary penalties in cases of rape and murder compared to other states. Several additional trends were also identified. Our experiments demonstrate that the proposed framework has the potential to identify the anomalies that could help in further understanding the causal factors, such as bias, that contribute to such anomalies and disparities. We also provide a structured dataset annotated by domain experts, which treasures sentencing-related information along with the verdict rationale from judgements of around 10,000 criminal cases adjudicated in the Trial court, High Court, and Supreme Court of India during 2000-2010. We make this dataset public to encourage further research

Abstract

biomedicine, but their size increasingly exceeds the on-board memory of modern GPUs. While GPUs offer substantial parallelism and memory bandwidth, many real-world inputs fit in CPU memory but not in GPU memory, making naive GPU-only methods inapplicable and unified-memory or zero-copy approaches unpredictable on irregular workloads. This thesis introduces the Parallel Heterogeneous External Memory (PHEM) model—a principled framework for processing graphs that exceed GPU memory while leveraging both CPU and GPU effectively. PHEM targets the regime where the full edge set must be streamed to the device, but the vertex set and modest auxiliary state fit on the GPU. The model partitions an input edge stream into a CPU portion and GPU batches and overlaps communication with computation so that the GPU processes batch i while batch i+1 is transferred. A tunable parameter controls the CPU–GPU work split, enabling adaptation to hardware and dataset characteristics. We present a generic algorithmic template with three customizable components—CPUCompute, GPUCompute, and GPUMerge—that maintains incremental state across batches and minimizes device-space footprint, requiring only one or two passes over the input. We instantiate the model for minimum spanning tree computation (PHEM-MST). On the GPU, we design a memory-efficient Boruvka variant on an edge-list representation that avoids symmetric dupli- ˚ cates and uses a two-phase atomic procedure to find best neighbors, followed by iterative shortcutting and filtering. In parallel, the CPU executes a filter-based Kruskal to prune non-MST edges from its partition. A final GPU-side merge combines the CPU and GPU forests; correctness follows from the optimal substructure of MST. We evaluate on eleven graphs spanning road networks, web crawls, social networks, synthetic generators, and biomedical literature—up to 11.6B edges—on two heterogeneous systems (NVIDIA L4, 24 GB; NVIDIA A100, 40 GB) paired with AMD EPYC CPUs. Against state-of-the-art CPU and GPU baselines (PBBS, ECL-MST) and a monolithic GPU variant, PHEM-MST processes all datasets, including those that exceed GPU memory for baselines, and achieves speedups up to 3.5×. Performance improves further on the A100, underscoring the role of device memory bandwidth. We analyze the impact of batch count and show dataset-specific optima that balance compute, contention, and transfer overheads; explicit batching with overlapped transfers yields more predictable performance than unified-memory mechanisms on irregular access patterns. Contributions include the PHEM model and reusable framework for heterogeneous, out-of-memory graph processing; an efficient PHEM-MST design and implementation; and a comprehensive empirical iii iv study with guidance for tuning and portability. We discuss limitations (parameter sensitivity, structureaware partitioning) and extensions (multi-GPU scaling, disk-backed execution). Beyond MST, the model naturally extends to BFS, connected components, and PageRank on graphs that outgrow GPU memory.

Abstract

Water quality monitoring of inland waterbodies is a global challenge due to limited monitoring infrastructure and availability of high-resolution hydrological data. Timely assessment and mitigation of water quality threats is constrained by limited data availability. To address this issue, remote sensing techniques have emerged as a robust tool for continuous real-time monitoring of inland water quality. In tropical regions, there has been a greater emphasis on remote sensing techniques for the continuous monitoring of water quality, mainly in eutrophic or hypereutrophic inland water bodies such as estuaries, rivers, and lakes, while reservoirs have received comparatively less attention. However, worldwide, tropical reservoirs are vulnerable to contamination risks due to their multifunctional roles, i.e., drinking, irrigation, and hydropower generation and long periods of storage times without adequate outflows which are critical for maintaining downstream water quality. Furthermore, the surrounding catchment influence contamination dynamics, as rapid changes in land use/land cover (LULC) can introduce sediments, nutrients, and heavy metals, contributing to elevated levels of physical, biological, and chemical characteristics of water quality parameters. Identifying the critical importance of monitoring tropical reservoirs, the present study focuses on employing remote sensing techniques to estimate water quality parameters, focusing on physio-biological characteristics, mainly in oligotrophic and mesotrophic reservoirs, and underscores the significance of catchment-related factors in influencing contamination levels. This approach is essential for maintaining optimal water quality levels, benefiting both local communities and the surrounding environment. The study selected two tropical reservoirs, Bhadra and Tungabhadra, situated within the same river system and experiencing similar climatic conditions but distinct landscape characteristics. The primarily aim is to develop a modeling framework to map and quantify the spatiotemporal variability of Chlorophyll-a (Chl-a), turbidity and surface water temperature (SWT), along with associated water spread for 2016-2021 using Sentinel 2 and Landsat 8 satellite datasets. The second aim is to examine interconnections among selected water quality parameters, to improve the accuracy of reservoir water quality assessments and identify potential correlations and interdependencies. The third aim is to comprehend how the landscape (land use/land cover (LULC)) and meteorological variables (rainfall, air temperature) of the contributing catchment influence contamination dynamics. The remote sensing-based results, validated with field-based observations measured from the Aquaread AP7000 instrument and secondary data include EOMAP (Earth Observation and Mapping) water quality derived datasets were employed to enhance accuracy and reliability. Chl-a, used as a proxy for nutrient contamination primarily from agricultural runoff, estimated using the Maximum Chlorophyll Index (MCI) derived from Sentinel 2 satellite data in the Google Earth Engine (GEE) platform. The results indicate a consistent and extensive distribution of Chl-a across the entire water surface of the Tungabhadra reservoir, while the Bhadra reservoir exhibited a more limited distribution. During the post-monsoon, increase in Chl-a spread in the Tungabhadra reservoir, primarily attributed to nutrient-rich water inflows from agricultural and urban areas. This notable rise linked to the harvesting of Kharif crops. Additionally, the discharge of untreated sewage further contributes to the degradation of water quality in the Tungabhadra reservoir. In contrast, the Bhadra reservoir, surrounded predominantly by forested areas, maintained water cleanliness and served as a riparian boundary. These findings demonstrate how LULC changes drive nutrient contamination in tropical reservoirs. Turbidity was considered as a proxy for sediments, detected and mapped using Sentinel-2 data. The Normalised Difference Turbidity Index (NDTI) values strongly correlated with EOMAP and field-based observations having values > 10 NTU (nephelometric turbidity unit) with R 2 = 0.81 and standard error estimate (SEE) of 5.3, respectively. In addition, the correlation between red band reflectance values with < 10 NTU, demonstrated a strong relation with an R 2 = 0.74, SEE = 1.7. Combining NDTI and red band reflectance allowed classification into low, moderate, and high turbidity zones. Further analyses indicated that diverse land use patterns, particularly high proportions of urban and agriculture area along with rainfall, significantly influenced the annual and seasonal turbidity variations. Surface water temperature (SWT) is a significant physical parameter affecting aquatic metabolism, nutrient cycling, and contaminant interaction was estimated using Landsat 8 Thermal Infrared (TIR) d

Abstract

While text-to-image (T2I) models have demonstrated remarkable improvements in spatial consistency and positioning of large scene components, they struggle with fine-grained structural control, particularly for anatomically complex objects with multiple interconnected parts. This limitation manifests as anatomical inconsistencies, missing or hallucinated components and incorrect part connectivity, critical failures for applications requiring precise structural fidelity. Recent developments in controlled image generation have produced models capable of high-fidelity synthesis. However, these models either rely on sophisticated conditioning inputs such as detailed pose skeletons or segmentation masks that require human expertise. We introduce PLATO, a novel two-stage framework that bridges this gap by enabling precise, partcontrolled object generation from simple, intuitive inputs: an object category and a list of constituent parts. The first stage employs PLayGen, our novel part layout generator built upon a Graph Convolutional Network Variational Autoencoder (GCN-VAE) architecture. PLayGen consists of a GCN Refinement Decoder that iteratively refines part placements through graph-based message passing, capturing inter-part spatial relationships during refinement. To address the challenge of training with nonintersecting bounding boxes and small parts, we introduce the Dynamic Margin IOU (DyI) loss, which dynamically adjusts margins based on box dimensions to ensure meaningful gradients throughout training. Additionally, our Differential Refinement loss scheme enables the model to learn both coarse positioning and fine-grained adjustments simultaneously. The overall loss formulation also has the effect of stabilizing the training process while improving structural coherence. In the second stage, PLayGen’s synthesized layout guides a custom-tuned ControlNet-based diffusion model through a novel visual conditioning format using color-coded inscribed ellipses combined with part-aware text prompts. This design enforces both spatial constraints and semantic consistency, resulting in anatomically accurate, high-fidelity object generations that faithfully contain precisely the user-specified parts. We observed that standard layout and image quality metrics do not sufficiently capture the nuances of structural composition. To address this gap for a rigorous structural consistency evaluation, we also propose novel part-level evaluation measures including Disconnection Percentage (measuring layout connectivity), Part-Layout IOU (quantifying spatial adherence) and Part-Layout F1 score (assessing part presence accuracy). Experiments on PASCAL-Parts and PartImageNet datasets demonstrate that PLATO significantly outperforms state-of-the-art approaches across both layout quality and structural controllability

Abstract

Automated segmentation of medical image volumes holds immense potential to significantly reduce the time and effort required from medical experts for annotation. However, leveraging machine learning for this task remains a formidable challenge due to variations in imaging modalities, the inherent complexity of medical images, and the limited availability of labeled patient data. While existing interactive segmentation methods and foundational models incorporate user-provided prompts to iteratively refine segmentation masks, they often fail to learn the continuity and inter-related information between consecutive slices in a 3D medical image volume. This limitation leads to inconsistencies, spatial discontinuities, and loss of anatomical coherence, ultimately affecting the reliability of segmentation results in clinical applications. The work proposes a novel interactive segmentation framework that dynamically updates model parameters during inference using a test-time training paradigm guided by user-provided scribbles. Unlike traditional approaches, our method preserves crucial spatial and contextual information from both previously processed slices within the same medical volume and the training dataset through a studentteacher learning mechanism. By leveraging sequential dependencies across slices, our approach ensures smoother and more anatomically consistent segmentation masks while integrating prior knowledge from the training distribution. We extensively evaluated our framework across diverse datasets, encompassing CT, MRI, and microscopic cell images, demonstrating its superior performance in both efficiency and accuracy. Our method significantly reduces user annotation time by a factor of 6.72× compared to manual annotation workflows and factor of 1.93× compared to state-of-the-art interactive segmentation methods. Furthermore, when benchmarked against foundational segmentation models, our framework achieves a Dice score of 0.9 within just 3–4 user interactions—substantially improving upon the 5–8 interactions required by existing models. This reduction in required interactions translates to a more streamlined and intuitive annotation process for volumetric CT and MRI scans. Additionally, our framework exhibits strong generalization capabilities, effectively segmenting unseen objects with minimal user guidance while maintaining spatial continuity across slices. Its ability to dynamically adapt during inference, integrate sequential information, and leverage interactive feedback makes it a highly effective tool for medical image analysis. To facilitate adoption and further research, we will publicly release the full source code, pre-trained models, and the developed annotation tool upon publication. This work underscores the transformative potential of our framework in enhancing the efficiency, accuracy, and consistency of medical image segmentation, addressing a critical need in the medical imaging field.