For past few weeks, I have been trying to teach body parts to my toddler. I started out by pointing to my ‘nose’ followed by saying ‘nose’. I did this consequently for few days before moving to other body parts and then gradually moving on to the game of finding body parts.
I would call out ‘Where is baby’s nose?’ Then asking my toddler to point to his nose. Surprisingly he would point to the right body parts most of the times. He usually has a problem with follow up questions and would get answers wrong if multiple or more questions followed the first question.
This game of finding body parts made me realise that progressively harder tasks are still a challenge for most algorithms. Recently Facebook decided to discontinue its virtual assistant M. Facebook found that when M would complete tasks, users asked for progressively harder tasks.
The team working on M soon realised that a fully automated assistant would have to do things far beyond the capabilities of existing machine learning technology. Human intelligence is better at understanding the nuances of natural language which means my toddler will have answers to all questions eventually.
As marketers we live in a world where proprietary algorithms decide which ads your consumers get to see, which leads have a higher chance of conversion, and how should your budget be allocated.
AI and Machine Learning are getting marketers closer to one of advertising’s most-sought goals: Relevance at Scale. This scale means as marketers we can customize campaigns in real time serving customers intent.
Google deployed machine learning extensively to reach and engage target audience for the launch of Pixel Phone. The team at Google used a new feature on Doubleclick tool called Custom Algorithm. Custom Algorithm makes it easier to customize the DoubleClick Bid Manager algorithm which allowed the team at Google to improve their strategy based on proprietary data and models.
By making use of historical data, the algorithm increases the likelihood that pixel ads are served to the most relevant, high-value audience based on the specific objectives. The tool uses machine learning to increase the number of viewable impressions bought on premium placements. The end results for Pixel were impressive. When compared to other Google campaigns that didn’t use the tool, impressions on premium inventory more than tripled, and viewable CPM fell 34%.
Instacart has built a machine-learning model to predict the most efficient sequence its shoppers could follow to select items at a store. The Walt Disney Co. is using language processing to trigger an audio soundtrack when you’re reading a story aloud to your child.
Deep Learning Framework
Google open-sourced Tensorflow, the companies software library for machine intelligence in 2015. Tensorflow is today among the most popular machine learning projects on GitHub. Tensorflow has found several use-cases including language translation, early detection of skin cancer and preventing blindness in diabetes
Similarly, Microsoft open-sourced Microsoft Cognitive Kit (formerly known as CNTK), and Baidu announced the launch of Paddle Paddle. If you are a developer, you have several options when it comes to choosing a deep learning framework.
Deep learning involves training artificial neural networks on lots of data and then getting them to make inferences about new data. For instance, an enterprising Japanese cucumber farmer trained a model with TensorFlow to sort cucumbers by size, shape, and other characteristics using the framework’s image recognition capabilities.
Getting Started with Deep Learning
If you plan to train yourself in Tensorflow or any other deep learning framework you should begin by learning Python programming language. If you understand Python but not machine learning, then you have some catching up to do.
A basic machine learning tutorial will cover concepts and algorithms related to machine learning, application, and how to evaluate the performance of machine learning algorithms. If you understand both Python and Machine Learning basics, then you can start by downloading the Tensorflow tutorial.
Today we have a very limited number of people who can create advanced machine learning models. Designing a custom ML model is a time intensive and complicated process. So how do you democratize the power of AI?
To close this gap, Google recently introduced Cloud Auto ML. Cloud Auto ML allows people with limited machine learning expertise to start building their own high-quality custom model. Cloud Auto ML currently provides support for less-skilled engineers to create custom ML models for image recognition.
To get web developers to start using machine learning, Google launched a new library called deeplearn.js. The library allows web developers to train a neural net right in your browser—locally on your device—without sending any images to a server. Teachable Machine is an AI experiment that showcases this new libraries capability.
How Can Deep Learning Help Marketers?
As marketers, we are already using deep learning frameworks. Machine learning is already used for sentiment analysis, predictive lead scoring, smart replies on chat, marketing automation etc.
If you plan to experiment with a deep learning framework, you could start by using regression analysis. Regression analysis helps in estimating the relationships between among variables. Regression Analysis can help marketers in the following ways:
- Analyzing marketing effectiveness – Marketing Mix Analysis (Linear Regression)
- Forecasting demand for a product/service – Demand Forecasting (Linear Regression)
- Estimating Customer Life Time Value (Lasso Regression)
- Increasing footfalls in a retail store by identifying the best combination of factors. (Multivariate Regression)
Linear regression is perhaps the most understood algorithm in statistics and machine learning. Beginners can start by building a linear regression model on a deep learning framework.
For social media marketers interested in designing social listening tools you can create a model for sentiment analysis using Facebook FastText an open-source library for building scalable solutions for text representation and classification.
Deep Learning for Marketing: Endless Possibilities
As marketers, if you want to move the bar when it comes to deep learning your approach has to be outward focused. To push the envelope forward, you need to collaborate with the marketing community because like anything else technology only gets better with human intervention.