[1]. The SLD conditions manifest as a deficit in processing language, spoken or written, that may manifest itself as a difficulty to comprehend, speak, read, write, spell, or to do mathematical calculations and includes such conditions as perceptual disabilities, dyslexia, dysgraphia, dyscalculia, dyspraxia, and developmental aphasia. SLD interferes with the normal learning process of the person. One-third of people with learning disabilities are estimated to also have attention-deficit hyperactivity disorder (ADHD). Further, it is estimated that nearly 5-15% of children struggle with Specific Learning disabilities (SLDs) in India [2].

The cognitive flexibility associated with SLDs can manifest itself in noteworthy talents, which include a multi-sensory lens for creative and lateral thinking, resulting in out-of-the-box solutions for problems. The untapped potential of SLDs causes high opportunity costs for the Nation’s progress. However, prevailing learning environments for SLDs create disparity in the education system, trigger divergence from the policy of ‘Learning for All (NEP-20)’, and depart from the provisions of ‘The Rights of Persons with Disabilities Act (2016)’. Feelings of isolation and a loss of interest in learning are often reflected in children with SLDs. Children with SLDs experience repeated failures and poor performance despite their continuous efforts and practice in learning [3]. At the same time, worldwide, the condition with SLDs has been exacerbated due to the COVID-19 pandemic when education delivery shifted online. According to global experts, “Future of Schools” is a hybrid model, where students will be both; on & off-campus.

Thus, strengthening online education delivery will be important and impacting. However, research has indicated that educators might not always be aware of their students’ attentional focus, and this may be particularly true for novice teachers [4]. The effort further increases when a single educator has to monitor the attention of the class at the individual level rather than the group level and across the entire class duration. Hence, technological tools that can improve and monitor the attention of children with SLDs can play a significant role in their inclusion during digital learning.

“EngageMe” aims to develop an intelligent platform that will offer personalized, monitored and evidence-based identification of attention levels among children with SLDs during digital learning. We will employ novel sensing technologies for multimodal behavioral analysis of the child’s online engagement using physiological, behavioral, and contextual information in a non-intrusive manner. Using Artificial Intelligence (AI) and Machine Learning (ML), we aim to better understand the cognitive state and affective processes behind attention and engagement during digital learning. Further, we will develop intelligent just-in-time and just-in-place interventions that can enhance the digital learning experience and better support emotional wellbeing among children with SLDs. EngageMe will help the special educators and pedagogues in reaching an objective and reliable assessment of the child’s attention level during online learning. Given the ongoing pandemic scenario, currently, we are collecting data using an online portal. One can head over to know their attention level by performing a sequence of three different simple and interactive psychological tasks. We look forward to building new collaborations with researchers, special educators, care facilities working with children with SLDs working around the country. Thanks to the support provided by iHub Anubhuti, we are excited to bring this project to fruition and look forward to enabling the untapped potential of children with SLDs.

About the Principal Investigator:
Dr. Jainendra Shukla leads the Human-Machine Interaction (HMI) Lab at Indraprastha Institute of Information Technology, Delhi (IIIT-Delhi). He is an Assistant Professor at the Department of Computer Science and Engineering in joint affiliation with the department of Human-Centered Design. He is also serving the Centre for Design and New Media as the head and is associated with Infosys Centre for Artificial Intelligence. He is experienced in Affective Computing, Human-Computer Interaction, and Social Robotics.

References:
[1] Singh, S., Sawani, V., Deokate, M., Panchal, S., Subramanyam, A. A., Shah, H. R., & Kamath, R. M. (2017). Specific learning disability: A 5-year study from India. Int J Contemp Pediatr, 4(3), 863-8.
[2] Ministry of Social Justice and Empowerment. Notification, 2018, Gazette of India (Extra-Ordinary) Department of Empowerment of Persons with Disabilities (Divyangjan) 2018. Jan 4, [Last accessed on 2021 Dec 12],
[3] Sahu, A., Patil, V., Sagar, R., & Bhargava, R. (2019). Psychiatric comorbidities in children with specific learning disorder-mixed type: A cross-sectional study. Journal of neurosciences in rural practice, 10(4), 617.
[4] Goldberg, P., Sümer, Ö., Stürmer, K., Wagner, W., Göllner, R., Gerjets, P., … & Trautwein, U. (2019). Attentive or not? Toward a machine learning approach to assessing students’ visible engagement in classroom instruction. Educational Psychology Review, 1-23.

Need for Summarization after Counselling
Unlike general clinical discussions, psychotherapy’s core symptoms are hard to distinguish, thus becoming a complex problem to summarize later. A structured counseling conversation may contain discussions about symptoms, history of mental health issues, or the discovery of the patient’s behavior. That’s why it is important and necessary to use online counseling conversation summarization that will directly help the therapist.

General Summarization Model:
Various speech recognition and dialogue summarization has been built with time to help people generate documents and reduce time. In the healthcare sector, various online medical conversation summarization platforms had also been made to provide solutions with patients’ medical data but due to lack of knowledge of such an online platform, it is hardly used by doctors and medical staff. And it is extremely important in mental healthcare to keep a record of the patients’ data and keep critical information like medical history.

ConSum Model : Counselling Conversation Summarization in Mental Healthcare
ConSum, an online Counselling Conversation Summarization Model, a psychotherapy intervention technique is a multifaceted conversation between a therapist and a patient.

Aseem Srivastava and Yash Kumar Atri, Ph.D. Students – IIIT Delhi, under the guidance of Dr. Tanmoy Chakraborty have been working on this project and developed this model which helps therapists directly in the counseling sessions. Here are some details related to the proposed ConSum model discussed above.

ConSum undergoes three independent modules:
1. To assess the presence of depressive symptoms, it filters utterances utilizing the Patient Health Questionnaire (PHQ-9). 2. Classification of essential counseling components that play a major role in summary generation. 3. Propose a problem-specific Mental Health Information Capture (MHIC) evaluation metric for counseling summaries.

The main focus of the Consum Model is to summarize the symptoms/reasons, routines, and patient discovery. It Summarizes the whole conversation in meaningful and relevant information with good grammar and linguistics. It makes sure that the summary is completely unbiased.

It’s remarkable to see the use of artificial intelligence and deep learning in the current state-of-the-art models. Researchers from LCS2 Lab, IIIT Delhi have used advanced deep learning methods including transformer-based models to build quality counseling summaries.

The comparative study shows that ConSum Model improves performance and generates cohesive, semantic, and coherent summaries. It comprehensively analyzes the generated summaries to investigate the capturing of psychotherapy elements (aka counseling components). Human and clinical evaluations of the summary show that ConSum generates quality summaries.

Mental health experts from Mpathic.ai validate the clinical acceptability of the ConSum summarization model. The ConSum application has been approved to be commercialized for therapists and other mental health experts worldwide. The first version of the application will be rolled out soon.

Teachers can also leverage this technology by automating their day-to-day classroom activities such as checking answer scripts, preparing mark sheets, and keeping a proper eye on academically challenged students. They can contextualize and personalize their lecture slides to make them more interactive and convey their insights in a meaningful way.

The applications are not limited to classroom settings and can also be deployed by the administrative services and support staff to make things easier and efficient. The segregated specialized divisions, such as student finance, academic support, hostel support, career support, can use these cognitive services to address the concerns of students and automate the notification process to keep them updated with any changes. This will save valuable time and resources and increase the overall efficiency of the response system.

Libraries can deploy this system to catalog their offline books, digital content, and research papers to enable faster search access for both students and faculty. Career counselors, which are often understaffed, can also enhance their role on campus by providing comprehensive service to students who seek their advice.

Cognitive computing and Artificial Intelligence are bound to become ubiquitous in the educational sector. Various services availed by different stakeholders of this field are set to utilize one or more features of these emerging technologies. Their proper usage can ramp up the educational standards of any college, university, or school to deliver advanced services that will not only increase student retention but also expand the academic horizon beyond their walls. Many academic communities are all set to install these changes for uplifting their intellectual structure.

The situation in the past two years has emphasized a need to have robust and scalable healthcare systems supported and complemented by technology. We have seen how technology has been harnessed to provide access to health infrastructure and medical resources, connecting citizens of the country to what may be called their digital health identities. Even before the Coronavirus crisis, there was a growing trend of healthcare and patient data becoming available digitally, ready for computation and analysis. When applied to such healthcare data, structured or unstructured data, Artificial intelligence has excellent use. Even though it is foreseen that artificial intelligence will not wholly replace human health professionals, it will undoubtedly play a huge role in assisting them in activities like screening for anomalies and diagnosis. In a significant way, Artificial Intelligence can use sophisticated algorithms to derive inferences from healthcare data. These inferences will help provide real-time medical procedures, navigate complications, etc.

Predictive modeling can help predict outcomes of treatments and help successfully walk through them. Robots are already being used in surgeries, and quite successfully at that. Diagnostic activities make good use of image processing, which helps process scores of diagnostic and medical testing images. Artificial intelligence is also being used to detect cancer at its earliest stages at this stage of diagnosis. Artificial intelligence is also being used in researching and developing new treatments and drugs. While the potential for AI to independently conduct all aspects of the drug development process is still limited, AI significantly aids in individual stages of the drug development process. It speeds up clinical trial processes by creating more efficient methods of subject recruitments. It also optimizes the process of drug selection by quickly eliminating those candidates most likely to fail clinical trials. Some real-world applications of artificial intelligence and machine learning in medicine include IBM Watson Genomics, a significant application of cognitive computing to generate insights from sequencing tumors and thus help develop specific treatments.

Another example that further emphasizes the utility of data-driven approaches using these novel concepts and AI and cognitive computing technologies is Google DeepMind. It is a novel move at creating timely response systems that alert healthcare providers about any anomaly or aberration in the patient’s condition so that they may provide immediate support to the patient, in many cases, which might be vital for the survival of the patient. The critical portion one looks at while considering the future of cognitive computing and artificial intelligence in healthcare is the ability of pattern detection, identification, and analysis. Understanding a patient and a human physician does and using data-driven methods to arrive at conclusions beneficial to the patient’s well-being again form the basis of the future of cognitive computing and artificial intelligence in healthcare. While there might be doubts about whether such cognitive computing solutions can fully replace human healthcare personnel, they can undoubtedly be utilized to assist, and complement the efforts of such personnel, now and in the future.

As of May 2022, over 4.7 crore cases are pending in different Indian courts (Data: The Hindu). Artificial Intelligence (AI) can go far beyond just automation in assisting the judges and lawyers. The theory of an AI agent (i.e., becoming better and better, the more it is practiced and the more data it is fed) matches with the way of working of a law system. AI-powered legal automation will be commonplace in the next five years to come. Artificial Intelligence can further aid the litigants, lawyers, and the judiciary through better document representation, document classification, discovering relevant documents, Online Dispute Resolution, Recovering Case Histories, Document Analysis, Legal Search, Strategizing the Case, Legal Summarization, Judgment Prediction, and much more.

The Government of India has taken active efforts to present all laws/statutes along with their amendments at indiacode.nic.in and all court judgments at judis.nic.in. Some noteworthy developments of digitization and the application of AI to the Indian Judicial System are LIMBS, E-Courts, SUVAAS, SCI-Interact, SUPACE, SPOTDRAFT, CASEMINE, CASEIQ, NEARLAW, and PRACTICELEAGUE.

Kanoon Sarathi:
The two camps of AI, symbolic approaches (rule-based) and the sub-symbolic (statistical and machine learning) approaches, have their own say and their own pros and cons. The hybrid of the two camps, called the neuro-symbolic AI is the way forward to Artificial General Intelligence. Kanoon Sarathi is built upon this hybrid approach where the human cognition meets the standalone Statistical AI capabilities. Kanoon Sarathi is a multilingual portal for the Indian Judicial System that offers a platform for legal data curation, management, and sharing. This knowledge management platform will facilitate the interpretability and explainability of the decision support and the reasoning procedures applied. Currently Kanoon Sarthi focuses upon three use cases:
1. Open data access for legal domain experts as well as to the common public to improve their involvement in understanding the judicial system in the form of Acts and Laws, Court Cases information, and so on 2. Finding similar cases based upon the context present in the court case documents. 3. Legal judgment prediction to help the judges decide the acceptance ratio of the appeal/petition with explainability.
Simply putting- “We are helping various people to give relevant insight from their perspective into the law related documents that are generating today.”

The Team:
Dr. Sarika Jain – Assistant Professor, National Institute of Technology, Kurukshetra is the Principal Investigator for this project. Her team includes one JRF, Ms Pooja Harde and several interns, namely Sudipto Ghosh, Dev Nirwal, Deepak Jaglan, Ankush Bisht, Abhinav Dubey, Tejas Mahajan, and Chirag Garg.

Milestones Achieved:
1. Presented and Under Publication with CEUR Workshop Proceedings (Scopus Indexed)
Title: Constructing a Knowledge Graph from Indian Legal Domain Corpus
Venue: International Workshop On Knowledge Graph Generation From Text (Text2KG) Co-located with the Extended Semantic Web Conference (ESWC 2022)
Team: Sarika Jain, Sudipto Ghosh, Nandana Mihindukulasooriya, Abhinav Dubey, Pooja Harde, Ankush Bisht

2. Presented and Under Publication with Springer LNEE Series (Scopus Indexed)
Title: Investigating the Similarity of Court Decisions
Venue: International Health Informatics Conference (IHIC 2022)
Team: Sarika Jain, Deepak Jaglan, Kapil Gupta

3. Presented and Under Publication with Springer LNEE Series (Scopus Indexed)
Title: NyOn: A Multilingual Modular Legal Ontology for Representing Court Judgements
Venue: The International Semantic Intelligence Conference ISIC 2022
Team: Sarika Jain, Pooja Harde, Nandana Mihindukulasooriya

The global legal service market has been growing at a compound annual growth rate (CAGR) of 3.4% since 2015, and reached a value of around $713.7 billion in 2020. As per a report by Statista, this value will reach $908.26 billion by 2025, growing at a rate of 4.9% from 2021 to 2025.

The field of law has remained largely under digitized and rather slow to adopt new technologies and tools for far too long. However, change is underway and Cognitive Computing will play a huge role in transforming the age-old traditional practices of the legal industry worldwide. Technology has the potential to revolutionize every aspect of the legal field, from law firms and the corporate legal field to courtroom operations and handling of the enormous number of documents involved.

Mammoth online legal data resources, such as LexisNexis and Practical Law, are constantly improving their search engines to help lawyers find material relevant to their cases faster and quicker. Lex Machina, an AI tool, also assists lawyers in creating a case strategy based on previous results in similar cases. Lawyers can also seek assistance in summarization and note-taking, saving considerable amounts of time.

Automation of documents is another innovation that can help with the qualms of documentation. Automation software is a way to assist users avoid the legal jargon of a document template. By presenting them with a questionnaire that collects pertinent data, it helps the user by simplifying the document generation process. The required data and elements are then automatically placed into the final document, which is generated by the system and subsequently provided to the user, based on the input provided. By getting rid of inefficiencies along the way in this manner, documents can now be made in minutes rather than days.

Simplification like this means that law, which was previously an industry with strong gate-keeping, becomes a lot more accessible to the common person. There have been attempts to make chatbots to help people learn about legal proceedings. Corporations, too, are using chatbots that enable both clients and lawyers. A Lawyer Bot, for example, is a software that can automate tasks typically performed by lawyers. These bots are excellent for increasing the speed of work and providing a better experience by allowing clients to self-serve online. For example, people in UK are using a chatbot called DoNotPay to dispute their parking tickets. The law firm BakerHostetler uses ROSS, an IBM Watson powered supercomputing software, to handle bankruptcy cases.

Clients often ask their counsel questions like “how likely will we win, if we go for a trial?” or “Should I compromise with the other party?” Chatbots may soon even provide data-backed answers to these. Machine learning models are being developed to predict the outcomes of pending cases, by taking as input the factual patterns of similar relevant cases. AI can access large amounts of past data and help lawyers answer such questions more accurately. A start up called Blue J Legal is developing an AI-powered legal prediction engine with an initial focus on tax laws.

Such efficiency and automation can also be introduced to contract review. After a contract is signed, overseeing and supervising it is usually a hassle. Especially in the case of large corporations which have countless pending contracts and counterparties spread across several divisions. Various NLP-powered solutions that derive and appraise vital facts across a firm’s core of contracts have begun to be produced, simplifying the firm’s business commitment nature for its stakeholders. They also assist the departments in staying updated with when the contracts need to be renewed. Seal Software and Kira Systems are two companies that develop such tools.

The legal market’s large scale presents a strong opportunity for value creation and is a market for creativity. Although, the introduction of these technologies may be slow, one can be excited because they will certainly renew the legal business in the near future.

How is an extraction attack carried out?
The process of model extraction is quite similar to knowledge distillation. Attackers collect a large number of unlabelled data samples, which are then sent to the victim model as queries. The victim model outputs a prediction for each query, which the thief treats as the ground truth label for the query. The prediction might range from the confidence scores (softmax probabilities) to just the hard label (class name). These query-output pairs serve as training data for the thief’s own model, allowing them to create a copy of the victim model, which we’ll call the thief model. The attacks are relatively cheaper than training a model from scratch and do not require extensive training and parameter tuning..

Who can try to steal a model and why?
○ A malicious competitor or adversary might steal a model to craft adversarial examples. These examples can be used to “break” the victim model and showcase its shortcomings.
○ The victim themselves can try to gauge the security level of their model by performing such attacks.
○ A thief can try to steal the model for monetary gain, ideally by rebranding it as their own model and exposing it via a much cheaper API

Our Project scope and future goals
○ Provide a tool/framework for performing such attacks on trained models
○ Coming up with new methods for performing attacks in more and more information-restricted settings
○ Building defences to prevent such attacks in the future, showcasing the efficiency of such defences via our tool
○ As these methods require a deep dive into the inner workings of ML/DL models, our work can also contribute to the explainability aspect.

Akshit Jindal, Ph.D. – IIIT-D under the mentorship of Dr. Vikram Goyal, Professor, IIIT-D is researching how an ML model can be efficiently attacked in the least informative setting and still be extracted to a good enough extent.