It is a honor for me to speak at the next conference “AI, Data Analytics
& Insights Summit – DACH”, 11th – 12th November 2021. I will meet 250 Senior Participants from DACH Region.
I will talk about Artificial Intelligence applied to Research and Development.
250 Senior Participants from DACH Region
Interactive Senior Level Online Meeting
The twenty-first century has ushered in a new age that is coined as data science and big data analytics. Data-driven scientific discovery is regarded as the fourth science paradigm. Data science has been a core driver of the new-generation science, technologies and economy, and is driving new researches, innovation, profession, applications and education across both disciplines and business domains.
There are many scientific and technical challenges associated with big data, ranging from data capture, creation, storage, search, sharing, modeling, representation, analysis, learning, visualization, explanation, and decision making. Among the many data characteristics and complexities to be addressed, I mention the hybridization of heterogeneous, multisource, hierarchical,
interactive, dynamic, multidimensional, and quality-poor data mixed with real-time business operations, strategic planning, decision-making, value creation, and future developments.
The field of data sciences and big data analytics have been evolving from statistics since half century ago to broad areas including but not limited to data and signal analytics, knowledge discovery, information retrieval, machine learning, statistics, optimization, computing, and data management. The literature defined areas of Data Science that requires in depth knowledge and pure research to be effective. By synergizing the three big areas—statistics, informatics and computing, data science has been spreading to essential and specific areas such as
It is a honor for me to speak at the next conference FUTURE Labs 2021. I will talk about Artificial Intelligence applied to Research and Development.
This is the world’s most diverse, stimulating, and exciting event for the future of all labs.
Start-ups, disruptors and innovators from academia, Biotech, Pharma, Chemicals, Food & Beverage, Agricultural, Materials, FMGC, Architects, Planners and many more are invited to help pave the way and share their experiences and vision for the lab of the future.
Featuring industry leaders sharing their views and expertise across 9 themes including: AI & Machine Learning, Digital Transformation, Lab Operations & Efficiency, Data Management and more. All conference sessions will be in English.
I am honored to attend the FUTURE LABS Live 2021. Wonderful experience with talents around the world to talk about digital transformation, data science, Artificial Intelligence, Machine Learning.
Labs, be they focused on research and development or on the creation of products and services, are the backbone of industry. The former turn money into knowledge while the latter turn knowledge back into money. Despite this, the potential of labs remains underutilised.
Technological, organisational, and cultural change remains a challenge across multiple industries. Transforming the lab and those who use it must happen now!
Future Labs LIVE is essential and timely. Held over two days, with over eighty global experts Future Labs will provide comprehensive and interactive coverage of the key issues and technologies.
Our vision is to create the world’s most important R&D lab event focusing on digitalisation, connectivity, and automation.
Future Labs unites leading stakeholders from across the entire laboratory eco-system to create insight and to promote collaboration. We guarantee to give you an experience that is engaging, collaborative and global.
With the fast development of sophisticated machine learning algorithms, artificial intelligence has been gradually penetrating a number of brand new fields with unprecedented speed.
One of the outstanding problems hampering further progress is the interpretability challenge.
This challenge arises when the models built by the machine learning algorithms are to be used by humans in their decision making, particularly when such decisions are subject to legal consequences and/or administrative audits.
For human decision makers operating in those circumstances, to accept the professional and legal responsibility ensuing from decisions assisted by machine learning, it is critical to comprehend the models.
For areas like the healthcare domain, business, crime prediction, etc., mistakes in these areas can be catastrophic. For instance, to develop safe self-driving cars we need to understand their rare but costly mistakes. Therefore, it is imperative to explain the learned representations, relationships between the inputs and the dependent variables and decisions made by these models. To trust the model, decision makers need to first understand the model’s behavior, and then evaluate and refine the model using their domain knowledge. One critical issue associated with future automated systems based on machine learning is its misalignment with the objectives of its stakeholders. That is, whether these systems really behave reliably in unforeseen situations. They may perform pretty well on test cases, but might do the wrong thing in deployment in the wild. Ironically, it could also be revealed later that they were doing the right things for the wrong reasons. Hence, interpretability plays a significant role in assisting us to reduce errors.
In the machine learning community, recently, interpretability is defined as “the ability to explain or to present in understandable terms to a human”.
Besides definition, a much harder task is to quantify and measure interpretability. Hence, the effort is extended beyond typical machine learning research into human-computer interaction. There are also other studies on aspects of interpretability such as the plausibility of models: the likeliness that a user accepts it as an explanation for a prediction.
For evaluation metrics, there are no mutually agreed standards. Sometimes the evaluation methods are only applicable to a specific model.
I present some general evaluation metrics: fidelity, comprehensibility and accuracy, which are frequently used by some state-of-art works.
Fidelity: It is not realistic for the interpretation model to be entirely faithful to the black-box model. Fidelity demands that the interpretation model’s prediction should match that of the black-box model as closely as possible. In other words, the interpretation model tries to mimic the behavior of the model itself on the instance being predicted.
Comprehensibility: Comprehensibility requires that the interpretation results are understandable to the users. When building an interpretation method, we should take into consideration the limitation of human cognition. For instance, decision trees involving thousands of nodes and decision rules having hundreds of levels of if-then conditions are not interpretable in this sense, although they are commonly regarded as inherently interpretable algorithms for textual representations.
Accuracy: Accuracy measures the performance of the interpretation model on the original training data used to train the black-box model to check if the interpretation model could outperform the black-box model. The measurements could be traditional evaluation metrics in machine learning such as accuracy score, AUC score, F1-score, etc.
One interpretation method is a visualization technique that interprets deep Convolutional Neural Networks (CNN) via meta-learning, named CNN-INTE . Compared to LIME which provides local interpretations for the entire model in specific regions of the feature space, this method provides global interpretation for any test instances on the hidden layers in the whole feature space.
The second interpretation method applies the Knowledge Distillation technique to distill Deep Neural Networks into decision trees in order to attain good performance and interpretability simultaneously.
To know more about these interpretations, please look at these research papers:
This powerful technique equips machines with the ability to learn how to learn, a leap forward that has the potential to revolutionize industries, from healthcare to finance
Meta-learning, in essence, is teaching machines to become better learners. It’s about going beyond traditional machine learning, where models are trained for specific tasks with vast amounts of data. With meta-learning, we aim to create models that can adapt quickly to new tasks with minimal data. Imagine a world where AI systems can learn from a handful of examples, just like humans. Meta-learning brings us closer to that reality. By exposing models to a diverse range of tasks during training, they acquire the ability to generalize and apply their knowledge to unseen tasks effectively.
Indeed, traditionally machine learning models require vast amounts of data to become proficient at specific tasks. However, in real-world scenarios, obtaining abundant labeled data is often impractical or costly. This is where meta-learning steps in, offering a remarkable departure from the norm. Meta-learning is the art of training models to learn how to learn. Instead of focusing solely on one task, these models are designed to generalize from a diverse set of tasks, enabling them to adapt swiftly to new challenges. This is akin to equipping AI with the ability to learn from its own learning experiences – a game-changer in the world of artificial intelligence.
At its core, meta-learning involves two key stages: meta-training and meta-testing.
During the meta-training phase, the AI models are exposed to multiple tasks or datasets. They learn not only how to perform these tasks but also how to learn from them. This involves capturing higher-level features or representations that are transferable across tasks.
In the meta-testing, the magic happens. When faced with a new, unseen task, the meta-trained model adapts rapidly. It leverages the knowledge gained during meta-training to make predictions or classifications, even when provided with minimal task-specific data.
Few-shot learning is a machine learning paradigm that focuses on training models to make accurate predictions or classifications when provided with very limited examples or “shots” of each class or category. In traditional machine learning, models often require a large amount of labeled data for training, but few-shot learning seeks to address scenarios where obtaining extensive labeled data is impractical or costly. In few-shot learning, the training dataset for each class or category typically contains only a few examples, often ranging from one to a few dozen samples. This contrasts with traditional machine learning, where hundreds or thousands of examples per class are common. Few-shot learning often leverages meta-learning techniques, where models are trained on a variety of tasks or categories. The goal is to enable the model to quickly adapt to new tasks with only a small amount of training data. Meta-learning algorithms aim to learn a good initialization or prior knowledge that facilitates rapid adaptation.
Few-shot learning has practical applications in various domains, such as computer vision (e.g., recognizing rare objects or faces with limited training samples), natural language processing (e.g., text classification with few labeled documents), and recommendation systems (e.g., suggesting products with minimal user interaction history). It addresses the challenge of making accurate predictions in scenarios where collecting abundant training data is difficult or expensive.
The implications of meta-learning are vast and promising. Here are a few areas where it’s making a significant impact:
– Healthcare: Meta-learning can aid in the rapid development of AI systems for disease diagnosis. With minimal patient data, models can become proficient at identifying various medical conditions.
– Finance: In the world of finance, where market conditions change rapidly, meta-learning enables AI-driven trading systems to adapt swiftly to new trends and scenarios.
– Education: Meta-learning has the potential to revolutionize personalized learning. AI tutors can understand individual student needs and adapt their teaching methods accordingly.
In the pharmaceutical industry, one of the major challenges is the rapid and cost-effective discovery of new drugs. Traditional drug discovery processes are time-consuming and expensive, often taking years and substantial financial resources. Identifying potential drug candidates with high efficacy and low toxicity is a complex task that requires extensive experimentation and data analysis.
Meta-learning offers a transformative solution to accelerate and optimize the drug discovery process in the pharmaceutical industry.
Meta-learning allows models to learn from a limited set of drug-related data, including chemical structures, biological properties, and historical drug development data. Meta-learned models can also predict the efficacy of potential drug candidates, enabling researchers to focus their efforts on compounds with higher chances of success. It helps optimize drug formulations by considering various factors such as drug stability, solubility, and bioavailability, leading to the development of more effective medications. It can identify existing drugs with potential applications in new therapeutic areas, saving both time and resources.
