More is better than less. Neuroscience for learning efficiently.

Is it better to learn more or to learn less? Should I read small amounts every day or read a large amount every day to remember what I have just learnt. I still remember my teachers saying, “the best way to learn is to learn little by little every day. Like little drops of water turn into an ocean, learning little by little can accumulate and turn into a vast sea of knowledge.” Nevertheless, latest scientific evidence from neuroscience study shows that, paradoxically, learning a large amount of information provides us with a better acquisition of the information we are trying to retain.

…..paradoxically, learning a large amount of information provides us with a better acquisition of the information we are trying to retain.

Before we dive into the results lets learn a little bit more about how learning actually works in the brain.

Learning fast and slow? And the two systems of learning

In 2008, Nobel laureate Daniel Kahneman came out with an influential book: Thinking fast and slow. This book mainly argued that in human decision making there seem to be two kinds of systems, one which makes fast intuitive decisions and the other which makes calculative but slow decisions. Recent work in human learning seems to indicate that human learning also involves two distinct neurocognitive systems. One which is immediate and efficient and the other slower but robust in the longer term.

The fast learning system or the Working Memory (Baddeley, 2012) enables quick single- trial learning but comes with two caveats. It has a limited storage capacity and a time limit. Due to this when using working memory, we can accurately remember a small amount of information for a short period of time. Physiologically working memory is located in the part of the brain called prefrontal cortex, the place where your forehead is.

At the same time, Reinforcement Learning helps in slower resilient reward-based learning by learning the values of your choices (Dayan and Daw, 2008). Physically, this learning mechanism is based in the part of the brain called Striatum. Striatum lies deep inside your brain. This part plays an important role in releasing dopamine which encodes ‘reward prediction error which helps in assigning values.

So what is the deal with learning more? Learning a lot more means your working memory, gets full. This will, in turn, allow your brain to make associations by using the reinforcement principles. When you learn you don’t just reach and absorb the information you also connect and reinforce those ideas with other ideas. What researchers found was when we learn in small bits, we use reinforcement learning but our working memory interacts with reinforcement learning. Things learnt in the Working Memory fades as time passes. So this might be a good place to store information in the short term but in the longer term, it will be useless.

So what’s the way around it? Well for one we know that working memory is capacity limited. So why not just use fill it up and allow the mind to completely rely on the reinforcement learning system? This is what the scientists tested in their experiment. Read on to see what they discovered.

Testing Method

To test if learning in low load or high load makes a difference the researchers designed a three-step information and retrieval task. Participants were allotted into two groups to learn associations in the set size of 3 (for low load) and in the set size of 6 (for high load). After learning the associations (first-step), the participants had to do an unrelated task (second-step), after which they were tested on the associations they had learnt previously. The logic of doing this was that because of the set size the working memory will be full and the time break in between would make sure that working memory is incapable of providing answers in the testing phase (third-step). If the answers were correct then it would mean they were definitely learnt through reinforcement learning.

The researchers then created a computer model to predict what would happen if working memory could interfere with learning. According to this model, performance for high load learning would be worse than low load learning. However, counterintuitively in the testing phase, the performance for the high load learning would be higher than performance in low load learning. This is because in the low load learning task the working memory interferes with reinforcement learning and thus inhibits robust learning. The chart below will help you visualise it.

Fig 1: Model predictions (adapted from Collins 2018)


Results

To see how the model fares in practice, researchers asked 49 university students to go through the three-step information and retrieval task. Students were divided into two groups, one practising and testing on low load and another practising and testing on the high load of learning. The results were similar to the one predicted by their initial model. The students who had learned in the low load (with set sizes 3) performed better in the learning phase where they had to immediately reproduce their learnt associations. These students’ reaction times for answers were also significantly higher than the students who learnt in high load. These results are definitely intuitive. When you have to remember something and then immediately reproduce then you will definitely find it easier to remember less information than more information.

The interesting result came in when the same students were tested about the same information after an unrelated memory task. Here the students who had learnt in high load performed better than those who learnt in low load with fewer sets. The reaction times for students from high load task were lower than their counterparts from low load task. This means that when you attempt to learn more information you might not only retain more information but you will always be able to recall that information quickly and efficiently.

Fig 2: Results of Behavioural Experiment (adapted from Collins 2018)


To further understand the computational process behind this learning, the researchers applied three models to explain the data gathered from the behavioural experiment. The Three models included one based purely on reinforcement learning, the other based on both reinforcement learning and working memory-based learning and the final one based on reinforcement learning and working memory based learning interacting with each other. Their results gave confirmation of the results above. The model with reinforcement learning and working memory based learning interacting with each other explained the results of the behaviour study almost perfectly.

As a robustness check, Collin repeated their study again on another lot of 42 university students to check if the results do hold. And voila! the results did hold across the two different groups of university students.

Where does this all lead us?

Going by this replicable scientific study we learn a few important features of our learning process. First, we do have some kind of distinct learning process. These learning processes have different implications for short-term learning, long-term learning and retention. More importantly, these learning processes interact with each other in ways that might be detrimental to our long-term performance. However, by giving our learning process more time we could overcome this deficit and would just reap the benefits for time to come.

So next time, if you are reading an important report for a presentation perhaps don’t just read the introduction or the summary. Read a few pages more than that and you will remember things you read more accurately!

Do comment below to tell us what you think about the results of the study! Also if you have any suggestions or advice to improve this blog article please do comment below, or email me at r.kumar.6@warwick.ac.uk.

Reference Article & Notes

Baddeley, A. (2012). Working memory: theories, models, and controversies. Annual review of psychology, 63, 1-29. Retrieved April 14, 2018 https://www.annualreviews.org/doi/full/10.1146/annurev-psych-120710-100422?url_ver=Z39.88-2003&amp= 

Collins, Anne GE (2018). “The Tortoise and the Hare: Interactions between Reinforcement Learning and Working Memory.” Journal of cognitive neuroscience Early Access, 1-12. Retrieved April 14, 2018 https://www.biorxiv.org/content/biorxiv/early/2017/12/15/234724.full.pdf

Dayan, P., & Daw, N. D. (2008). Decision theory, reinforcement learning, and the brain. Cognitive, Affective, & Behavioral Neuroscience, 8(4), 429-453. Retrieved April 14, 2018 https://link.springer.com/content/pdf/10.3758/CABN.8.4.429.pdf

Note: All images are created by the author by adapting images from the reference article. This was done for visual accessibility.

All the images follow the creative commons licensing policy as required for blogs

Leave a Reply

Your email address will not be published. Required fields are marked *