Federated Learning (FL) offers a promising approach for training clinical AI models without centralizing sensitive patient data. However, its real-world adoption is hindered by challenges related to privacy, resource constraints, and compliance. Existing Differential Privacy (DP) approaches often apply uniform noise, which disproportionately degrades model performance, even among well-compliant institutions. In this work, we propose a novel compliance-aware FL framework that enhances DP by adaptively adjusting noise based on quantifiable client compliance scores. Additionally, we introduce a compliance scoring tool based on key healthcare and security standards to promote secure, inclusive, and equitable participation across diverse clinical settings. Extensive experiments on public datasets demonstrate that integrating under-resourced, less compliant clinics with highly regulated institutions yields accuracy improvements of up to 15% over traditional FL. This work advances FL by balancing privacy, compliance, and performance, making it a viable solution for real-world clinical workflows in global healthcare.

Digitization of scanned receipts aims to extract text from receipt images and save it into structured documents. This is usually split into two sub-tasks: text localization and optical character recognition (OCR). Most existing OCR models only focus on the cropped text instance images, which require the bounding box information provided by a text region detection model. Introducing an additional detector to identify the text instance images in advance adds complexity, however instance-level OCR models have very low accuracy when processing the whole image for the document-level OCR, such as receipt images containing multiple text lines arranged in various layouts. To this end, we propose a localization-free document-level OCR model for transcribing all the characters in a receipt image into an ordered sequence end-to-end. Specifically, we finetune the pretrained instance-level model TrOCR with randomly cropped image chunks, and gradually increase the image chunk size to generalize the recognition ability from instance images to full-page images. In our experiments on the SROIE receipt OCR dataset, the model finetuned with our strategy achieved 64.4 F1-score and a 22.8% character error rate (CER), respectively, which outperforms the baseline results with 48.5 F1-score and 50.6% CER. The best model, which splits the full image into 15 equally sized chunks, gives 87.8 F1-score and 4.98% CER with minimal additional pre or post-processing of the output. Moreover, the characters in the generated document-level sequences are arranged in the reading order, which is practical for real-world applications.

An important class of problems can be cast as inference in noisy(cid:173) OR Bayesian networks, where the binary state of each variable is a logical OR of noisy versions of the states of the variable's par(cid:173) ents. For example, in medical diagnosis, the presence of a symptom can be expressed as a noisy-OR of the diseases that may cause the symptom - on some occasions, a disease may fail to activate the symptom. Inference in richly-connected noisy-OR networks is in(cid:173) tractable, but approximate methods (e .g., variational techniques) are showing increasing promise as practical solutions. One prob(cid:173) lem with most approximations is that they tend to concentrate on a relatively small number of modes in the true posterior, ig(cid:173) noring other plausible configurations of the hidden variables. We introduce a new sequential variational method for bipartite noisy(cid:173) OR networks, that favors including all modes of the true posterior and models the posterior distribution as a tree.

Histone modifications play a critical role in gene regulation. Consequently, predicting gene expression from histone modification signals is a highly motivated problem in epigenetics. We build upon the work of DeepChrome by Singh et al. (2016), who trained classifiers that map histone modification signals to gene expression. We present a novel visualization technique for providing insight into combinatorial relationships among histone modifications for gene regulation that uses a generative adversarial network to generate histone modification signals. We also explore and compare various architectural changes, with results suggesting that the 645k-parameter convolutional neural network from DeepChrome has the same predictive power as a 12-parameter linear network. Results from cross-cell prediction experiments, where the model is trained and tested on datasets of varying sizes, cell-types, and correlations, suggest the relationship between histone modification signals and gene expression is independent of cell type. We release our PyTorch re-implementation of DeepChrome on GitHub \footnote{\url{github.com/ssss1029/gene_expression_294}}.\parfillskip=0pt

Artificial intelligence (AI) must be inclusive to reach its potential. AI applications that solve problems for a small segment of the population will fail to achieve widespread adoption. So, it's important that AI applications be designed and prepared with data that reflects as many segments of the global population as possible. Many moving parts need to be managed well to do that, and one of them is language. The more languages an AI application can handle, the more inclusive it is.

I've recently published an article explaining why the field of artificial intelligence could greatly benefit from the approaches of design fields. I believe that involving different stakeholders early on in AI-based projects is the most effective technique to battle the various kinds of biases and shortcomings embedded within AI systems. By starting at the very beginning, involved stakeholders and their insights can help shed light on inequitable processes of design, on systemic biases buried in data-sets and how they can disadvantage different groups of people, on use-cases and experiences that might otherwise be overlooked, and on potential consequences and implications that even the most rigorous testing might not capture. Instead of creating overly-specific, bespoke solutions tailored to a specific project, or trying to focus on completely eradicating one of those problems, my approach is to bring the voices that matter into the design process and let them help the expert team navigate all these challenges. By providing a more generalized toolkit and methodology for supporting these stakeholders, different problems in different projects and during different phases can all be addressed in whatever way is needed.

The Computer Science Department's new course focusing on issues of justice, equity, diversity and inclusion in computer science and society got its start when a group of graduate students decided to create the training they wished they had received. And after hundreds of hours of work by 15 Ph.D. students --pilot programs, countless conversations with faculty and students, data gathering, and developing and tweaking course material -- CS-JEDI: Justice, Equity, Diversity and Inclusion is now a required part of the curriculum for incoming Ph.D. students in computer science. It's also being looked at as a model by both other departments in the School of Computer Science and universities elsewhere. The course was created and taught by Abhinav Adduri, Valerie Chen, Judeth Choi, Bailey Flanigan, Paul Göelz, Anson Kahng, Pallavi Koppol, Ananya Joshi, Tabitha Lee, Sara McAllister, Samantha Reig, Ziv Scully, Catalina Vajiac, Alex Wang and Josh Williams -- all doctoral candidates in SCS who represent nearly every department in the school. The team received Carnegie Mellon University's 2022 Graduate Student Service Award and will be honored during the Celebration of Education Award Ceremony on Thursday, April 28.

Smart cities are supposed to represent the pinnacle of technological and human advancement. They certainly deliver on that promise from a technological standpoint. Smart cities employ connected IoT networks, AI, computer vision, NLP, blockchain and similar other technologies and applications to bolster urban computing, which is utilized to optimize a variety of functions in law enforcement, healthcare, traffic management, supply chain management and countless other areas. As human advancement is more ideological than physical, measuring it comes down to a single metric--the level of equity and inclusivity in smart cities. Essentially, these factors are down to how well smart city administrators can reduce digital exclusivity, eliminate algorithmic discrimination and increase citizen engagement. Addressing the issues related to data integrity and bias in AI can resolve a majority of inclusivity problems and meet the above-mentioned objectives.

Artificial Intelligence (AI) is appreciated for its potential to solve some of the biggest challenges that mankind is faced with — from drug discovery to climate change to poverty reduction and beyond. While it has made its way into many daily consumer uses, the lack of widespread enterprise applications is often cited by critics as a reality check over the hype cycle. What is certain in the coming year and beyond is that AI will continue to push the boundaries of what is possible in both consumer and business. Companies are already automating mind-numbing repetitive workplace tasks, empowering employees to focus on higher-value, creative problem-solving. A McKinsey survey showed that enterprise adoption was up 6 per cent from the previous year to 56 per cent in 2021. AI’s future in 2022 and the years to come can be looked at through the lenses of discovery, democratisation and de-risking. Discovering new frontiers The past 18 months challenged businesses on what can be accomplished almost overnight with the aid of technology. Digitally native firms, with AI embedded at the core of everything from architecture to operations, showed how to solve the most relevant problems whether it was to fix supply chain disruptions or support remote operations. Traditional businesses will learn from this, adopting an AI-first approach to solve those challenges that are relevant and feasible to them. To identify the most suitable use cases, organisations will need to be creative, curious and collaborative. Finding the right problems to solve will define the success of AI adoption. Additionally, organisation-wide data literacy will transform problem solving and value capture, which will cut costs, generate new revenues streams and make businesses more competitive. Recent advances in Quantum Computing will lead to the development of the next generation of algorithms and applications. This convergence with AI shows immense potential for the acceleration of Machine Learning (ML) and deep learning, resulting in new discoveries and use cases. ​​​​​​​ By the people, of the people, for the people In 2022, AI’s journey will not only be towards being more common, but also being more strategic. The bolt-on approach that has been common in the infancy stages, will be replaced by a deeper and wider adoption, becoming essential to the entire technology stack. AI’s power will be leveraged to rethink and rebuild services, products, business models and entire approaches, as enterprises move from an experimentative to a bolder implementation mode. Access to AI/ML models will also trickle down beyond the deep pockets of Big Tech and large enterprises to midsize businesses, courtesy the availability of more off-the-shelf and reusable assets in AI marketplaces. From unknown devil to responsible angel As AI increasingly permeates into human territory, the more scrutiny there will be around its decision making. Biased data that feeds AI systems is an almost universal ailment, further fuelling the mistrust against it. The call for transparency and ethics by design, is leading to the rise of AI that is at once explainable, ethical, auditable and even humble (when it knows that it is not sure about the right answer) — essentially AI that is responsible. IANS

Artificial intelligence (AI) is on the cusp of becoming democratic, inclusive, and useful to people living in under-served places in some very exciting ways. We believe an approach we call Mindful AI can help AI realize its potential to be more inclusive and human-centered, too. McKinsey's James Manyika interviewed Kevin to discuss concepts related to Kevin's recently published book, Reprogramming the American Dream: From Rural America to Silicon Valley – Making AI Serve Us All. The book draws on Kevin's personal experiences to show how AI can become more inclusive by helping people who live in under-served areas ranging from rural towns to working-class communities. For instance, as reported in The Wall Street Journal, Microsoft's FarmBeats program uses AI to improve farming.