You’re a Machine Learning professional who needs to communicate better. What’s the first step?

Importante entender sua audiência e conseguir explicar os conceitos de forma simples, sem usar muitos termos técnicos, a depender do nível de expertise deles em ML.

Engage them by asking open-ended questions to gauge their understanding and interests in real-time. This approach fosters an interactive learning environment, making complex ML concepts more accessible. By adapting in response to their feedback, you can fine-tune your message to ensure clarity and relevance.

?? Tailor your communication style and technical depth based on the audience's familiarity with Machine Learning concepts - are they technical? do they understand the data? etc are good questions to ask. ?? Consider cultural and professional backgrounds that may influence how your message is received and understood.

To effectively convey your message, understand your audience's interests and background. Put yourself in their shoes and consider what they should take away. Tailor your message to the appropriate depth and practice explaining concepts to diverse audiences for feedback. Use the advice of being able to explain complex topics in one sentence to someone like your grandma. This ensures you fully understand and can adapt the concept to any skill level.

I think following XAI or explainable AI like we do at Takara.ai is important not only so that we can explain our models and applications in digestible terms for the public but also because it establishes a culture where we strive to understand more about how these systems work and don’t end up in a place like Google where they don’t understand how the YouTube algorithm works and seemingly have no interest in finding out. This way we can augment our knowledge to any audience, technical or not.

In the context of the example, while it is necessary to avoid unnecessary details, it is also essential to provide an unbiased analysis of the results. It may be tempting to hide away unfavourable results and emphasise on positive aspects of the report, but that may lead to poor business decisions. The best approach is to: 1. Provide the pros and cons of the data, model and results. 2. Prepare and present a plan to tackle the identified challenges.

? Clearly define the goal of your communication, whether it's to inform, persuade, or collaborate. ?? Tailor your message to align with your audience's level of technical understanding.

?? Tailor your message for the platform, whether it's a technical report, email, or presentation, to ensure clarity and engagement. ?? Adjust the technical level of your communication based on the audience's expertise, from laypersons to tech-savvy peers.

For a ML researcher the only medium that matters for communication is the ML Application. Please remember that you will be giving a demo to "human" beings for a technology which is supposed to mimic "human" behaviour. Therefore, any human being, can judge you instantly and no one will be interested in the complex formulaes and neural network behind it. Just work hard before the demo and let the app do the talking. People will know when they are amazed especially if they have to judge something which they are so familiar with - human-like thinking.

Measuring and improving communication skills involves recognizing various feedback sources. Metrics like views, comments, and shares gauge impact in written content, while surveys and audience engagement assess effectiveness in speaking engagements. Seeking feedback from peers, mentors, and audience members provides valuable insights for refinement. Recognizing strengths and weaknesses enables targeted skill development. Embracing feedback cultivates continuous improvement, ensuring effective communication in diverse settings.

Mentioning your resources is very important in today's world. In case it is tedious to put stars in superscript everywhere, at least compile everything and put it at one place towards the end of the presentation.

Know yourself, but more importantly, know how to rein yourself in. As data scientists, we get carried away by the math. We also feel the need to explain everything in great detail, to show how much we know. Having been on both sides of the table I can say with certainty that the audience trusts you to have done the due diligence around the technique. They are more interested in understanding your point of view on the matter. The math is just the means, it's the logical thinking process you took to arrive to the end that matters more.

Speaking from my experience as an ML scientist and an expert in science communication, I can say that tailoring your language and ensuring you can communicate with both technical and non-technical audiences is important. For instance, when speaking to data scientists or engineers, it's okay to use technical jargon like "neural networks," "overfitting," or "gradient descent." However, when addressing a non-technical audience, such as business stakeholders or the general public, it's advisable to replace technical terms with simpler language. For example, instead of "convolutional neural networks," you might say, a type of computer program that's really good at recognizing patterns in images, similar to how our brains work.

Here are few things I try to do while communicating machine learning work: 1. Numbers are universal language - Use measurable KPIs and present KPIs. KPIs will differ based on the audience (like for technical audience it could be RMSE, ROC, or confusion matrix but for business audience it could be ROI or cost savings) 2. Establish a baseline - Always use a baseline to show improvements. The baseline could be a linear model or correlation for a technical audience and a before/after comparison for business audience. 3. Visuals are better than words - It is a cliché but a picture really is worth a thousand words. Use images (or graphs) to communicate trends.

The first step to improve communication as a Machine Learning professional is to understand the audience. Tailoring the message to suit their level of expertise and interests fosters better engagement and comprehension. Additionally, practicing clear and concise articulation of complex concepts helps convey ideas effectively. Moreover, actively soliciting feedback and actively listening to understand concerns aids in refining communication strategies. Lastly, incorporating visual aids and real-world examples can enhance understanding and retention of information.

When engaging with fellow data scientists, it's crucial to clearly articulate the methodology employed to address issues, including the operational mechanics and the technical metrics for assessing the efficacy of the method or model. In contrast, when interfacing with stakeholders or the business team, pinpointing the business metrics that resonate with their goals for AI-driven problem-solving is key. Presenting data in the context of these business metrics enables stakeholders to grasp the tangible benefits of implementing AI solutions.

Personally based in my experience in consulting and as end-business user 1) Know your audience: Consider their concerns i.e., marketing, sales, operations, etc. What do they care about i.e. the KPI, metrics etc. This is something that can be learned over a few weeks to months at any position. 2) Know the purpose or goal: What is the goal of the work you are presenting. Improve, reduce, re-calibrate etc. If looking for a buy-in show the impact by different perspective there could be none in general or it is marginal. Obviously visual aids + numbers help people understand. Final pointer which took a few meetings is the story line. A story is easy to understand.