A5100760604- Adversarial Robustness in Machine Learning
- Advanced Neural Network Applications
- Explainable Artificial Intelligence (XAI)
- Parallel Computing and Optimization Techniques
- Anomaly Detection Techniques and Applications
- Formal Methods in Verification
- Neural Networks and Applications
- Natural Language Processing Techniques
- Image Enhancement Techniques
- Network Security and Intrusion Detection
- Security and Verification in Computing
- Machine Learning and Algorithms
- Topic Modeling
- Model Reduction and Neural Networks
- Fault Detection and Control Systems
- Logic, programming, and type systems
- Image and Signal Denoising Methods
- Radiomics and Machine Learning in Medical Imaging
- Software Testing and Debugging Techniques
- Bone fractures and treatments
- 3D Shape Modeling and Analysis
- Medical Image Segmentation Techniques
- 3D Surveying and Cultural Heritage
- Advanced Image Fusion Techniques
- Algorithms and Data Compression
Goa Medical College
2025
University of Illinois Urbana-Champaign
2021-2024
Chaudhary Charan Singh Haryana Agricultural University
2023-2024
Chandigarh University
2024
All India Institute of Medical Sciences
2023-2024
Columbia University Irving Medical Center
2023
GNA University
2023
ETH Zurich
2015-2023
Indian Institute of Technology Delhi
2023
Clinical and Laboratory Standards Institute
2023
We present a novel method for scalable and precise certification of deep neural networks. The key technical insight behind our approach is new abstract domain which combines floating point polyhedra with intervals equipped transformers specifically tailored to the setting Concretely, we introduce affine transforms, rectified linear unit (ReLU), sigmoid, tanh, maxpool functions. implemented in system called DeepPoly evaluated it extensively on range datasets, architectures (including defended...
Formal verification of neural networks is critical for their safe adoption in real-world applications. However, designing a precise and scalable verifier which can handle different activation functions, realistic network architectures relevant specifications remains an open difficult challenge. In this paper, we take major step forward addressing challenge present new framework, called PRIMA. PRIMA both (i) general: it handles any non-linear function, (ii) precise: computes convex...
Numerical abstract domains are an important ingredient of modern static analyzers used for verifying critical program properties (e.g., absence buffer overflow or memory safety). Among the many numerical introduced over years, Polyhedra is most expressive one, but also expensive: it has worst-case exponential space and time complexity. As a consequence, analysis with domain thought to be impractical when applied large scale, real world programs.
The use of neural networks in safety-critical computer vision systems calls for their robustness certification against natural geometric transformations (e.g., rotation, scaling). However, current methods target mostly norm-based pixel perturbations and cannot certify transformations. In this work, we propose a new method to compute sound asymptotically optimal linear relaxations any composition Our is based on novel combination sampling optimization. We implemented the system called DeepG...
Seppo Enarvi, Marilisa Amoia, Miguel Del-Agua Teba, Brian Delaney, Frank Diehl, Stefan Hahn, Kristina Harris, Liam McGrath, Yue Pan, Joel Pinto, Luca Rubini, Ruiz, Gagandeep Singh, Fabian Stemmer, Weiyi Sun, Paul Vozila, Thomas Lin, Ranjani Ramamurthy. Proceedings of the First Workshop on Natural Language Processing for Medical Conversations. 2020.
Code generation, symbolic math reasoning, and other tasks require LLMs to produce outputs that are both syntactically semantically correct. Constrained LLM generation is a promising direction enforce adherence formal grammar, but prior works have empirically observed strict enforcement of constraints often diminishes the reasoning capabilities LLMs. In this work, we first provide theoretical explanation for why constraining very restrictive grammars only allow valid final answers reduces...
682 Background: Bladder cancer is the sixth most common malignancy in United States. Recent advances treatment paradigm have changed landscape of bladder cancer. Patient-reported outcome measures (PROMs) are symptom burden and health related quality life (HRQoL) that come directly from patient, without clinician interpretation. PROMs can be intervention and/or a trial. We aimed to evaluate trends usage PROs for patients with muscle invasive (MIBC) locally advanced or metastatic (mBC) as...
Hybrid storage systems (HSS) combine multiple devices with diverse characteristics to achieve high performance and capacity at low cost. The of an HSS highly depends on the effectiveness two key policies: (1) data-placement policy, which determines best-fit device for incoming data, (2) data-migration rearranges stored data across sustain performance. Prior works focus improving only placement or migration in HSS, leads sub-optimal Unfortunately, no prior work tries optimize both policies...
We introduce MedicalSum, a transformer-based sequence-to-sequence architecture for summarizing medical conversations by integrating domain knowledge from the Unified Medical Language System (UMLS). The novel augmentation is performed in three ways: (i) introducing guidance signal that consists of words input sequence, (ii) leveraging semantic type UMLS to create clinically meaningful embeddings, and (iii) making use weighted loss function provides stronger incentive model correctly predict...
In this paper, a comprehensive energy function is used to formulate the three most popular objective functions: Kapur' s, Otsu and Tsalli' s functions for performing effective multilevel color image thresholding. These new based criterions are further combined with proficient search capability of swarm algorithms improve efficiency robustness. The proposed thresholding approach accurately determines optimal threshold values by using generated curve, acutely distinguishes different objects...
A critical step of genome sequence analysis is the mapping sequenced DNA fragments (i.e., reads) collected from an individual to a known linear reference sequence-to-sequence mapping). Recent works replace with graph-based representation genome, which captures genetic variations and diversity across many individuals in population. Mapping reads sequence-to-graph mapping) results notable quality improvements analysis. Unfortunately, while well studied available tools accelerators, more...
Numerical abstract domains are a fundamental component in modern static program analysis and used wide range of scenarios (e.g. computing array bounds, disjointness, etc). However, with these can be very expensive, deeply affecting the scalability practical applicability analysis. Hence, it is critical to ensure that made highly efficient. In this work, we present complete approach for optimizing performance Octagon numerical domain, domain shown particularly effective practice. Our...