Suggestions
Ankit S.
LLM Arch Assoc Director@Accenture | Ph.D. Candidate at CMU LTI | Deep Learning | Machine Learning | AI | Computer Science | Past: Research Intern@Bosch, Google, MERL. DL Scientist@ReviveMed, Verification Engineer@ARM
Professional Background
Ankit S. is an ambitious and innovative professional in the fields of Artificial Intelligence (AI), Augmented Reality (AR), Virtual Reality (VR), and Language Learning Models (LLM). His passion for developing cutting-edge solutions has positioned him as a thought leader in Generative AI and AR/VR applications. With a career that spans numerous prestigious organizations, including notable tech giants such as Google and Bosch, Ankit has proven himself to be an invaluable asset across the tech industry.
In his previous role as LLM Associate Director at Accenture, Ankit contributed significantly to the development of AI-driven strategies and implementations, further honing his expertise in state-of-the-art technologies. His robust professional experience is complemented by multiple research internships at leading firms like Mitsubishi Electric Research Laboratories and Bosch USA, where he developed groundbreaking solutions in machine learning and AI.
Ankit's time at Carnegie Mellon University, where he served not only as a Doctoral candidate but also as a Graduate Research Assistant and Teaching Assistant, has endowed him with deep academic knowledge, particularly in the area of language technologies.
Education and Achievements
Ankit's educational journey is both impressive and extensive. He holds a Doctor of Philosophy (PhD) in Language Technologies and a Master of Science (MS) in Computer Science, both from Carnegie Mellon University — a globally recognized leader in technology education. His undergraduate studies culminated in a Bachelor of Technology (BTech) in Electrical, Electronics, and Communications Engineering from the National Institute of Technology Karnataka, one of India's premier engineering institutions.
In pursuit of continuous learning and professional growth, Ankit has also completed several courses through Udacity, including an introduction to self-driving cars, AI programming with Python, and Natural Language Processing. These courses have deepened his understanding of AI technologies and enhanced his capacity to innovate in his field.
Among Ankit's many accolades is the prestigious Gandhian Young Technology Innovation Award, awarded for his project titled "Never Ending Learning of Sound." This accolade is a national recognition celebrating outstanding contributions to technology innovation and reflects Ankit's dedication to leveraging technology for social good. Additionally, he is a multiple Bravo award recipient for his exceptional performance as a Verification Engineer at ARM, where he led the development of a universal test case suite that is utilized across various teams globally.
Core Skills and Areas of Expertise
Ankit possesses a mastery of numerous technical proficiencies essential for success in AI and related fields. Key skills include:
- Proficient programming in Python (having completed over 1600 LeetCode challenges), C, and C++
- Advanced understanding of machine learning, deep learning frameworks, and computer science principles
- Expertise in Natural Language Processing (NLP), vision systems, audio understanding, and sound event detection & classification
His areas of expertise reflect his commitment to ongoing professional development in emerging technologies. Ankit has cultivated knowledge in multimodal understanding, unsupervised and supervised learning methodologies, and advanced classification systems such as weakly and webly supervised learning techniques. His efforts in sound event detection and classification showcase his capabilities in innovating solutions with high accuracy, crucial for applications in various sectors, including safety and security.
Resilience and Drive
Ankit embodies resilience and drive in his career path, often drawing motivation from Winston Churchill's famous remark: "Success is going from failure to failure without losing your enthusiasm." This mindset has empowered him to tackle challenges and set benchmarks for excellence in technology development. Ankit’s resilience is evident in his proactive approach to professional growth, consistently seeking new learning opportunities and embracing collaborative endeavors.
Notable Projects and Contributions
Throughout his career, Ankit has made strides in addressing critical societal challenges through technology. One of his most noteworthy projects involved developing a cutting-edge AI framework for gun-type detection through acoustic analysis of gunshots. This pioneering work in leveraging AI to combat gun violence has achieved over 98% accuracy in gunshot detection and significantly contributes to public safety initiatives.
Additionally, Ankit has spearheaded advancements in multi-label classification systems for sound events, utilizing advanced supervised learning techniques. His contributions to audio understanding not only reflect technical prowess but also a commitment to enhancing the state of research in these domains.
Commitment t
Highlights
Our paper: Understanding and Mitigating the Label Noise in Pre-training on Downstream Tasks - https://lnkd.in/eAdcWvfG made it to ICLR 2024 as a spotlight paper. Authored with Hao Chen, Jindong Wang, Ran Tao, Hongxin Wei, Xing Xie, Masashi Sugiyama, Bhiksha Raj.
Key insights:
- Your foundation models are not clean since the (black-box) pre-training data are noisy! We propose a new research topic: Noisy Model Learning!
- Influence: The label noise in pre-training data has both benevolent and malignant influence on downstream tasks.
- Analysis: The label noise in pre-training shapes the feature space significantly of the pre-trained model.
- Mitigation: We design a simple black-box fine-tuning algorithm to reshape the pre-trained feature space, reducing the influence of noisy pre-training data and boost the performance of downstream tasks.
- Label noise in pre-training data can have varying effects on model generalization, improving in-domain transfer performance but consistently deteriorating out-of-domain performance.
- The presence of label noise in pre-training shapes the feature space differently, impacting the model's ability to generalize.
- The NMTune method is introduced as a lightweight black-box tuning approach to mitigate the adverse effects of label noise and enhance model generalization on in- and out-of-domain tasks.
- The paper introduces a new research direction termed Noisy Model Learning (NML), which is particularly significant in the era of large foundation models where downstream users often have only access to the model weights or APIs
