Bradley C. Love

Introduction Sutton’s Bitter Lesson (Sutton, 2019) argues that machine learning breakthroughs, like AlphaGo, BERT, and large-scale vision models, rely on general, computation-driven methods that prioritize learning from data over human-crafted priors. Large language models (LLMs) based on transformer architectures exemplify this trend, scaling effectively with data and compute. Yet, positional embeddings—a key transformer component—seem to challenge this philosophy. Most embedding schemes are fixed, not learned, and embed a...

Probability 101: Does Order Matter in Sequences? Ever tried reading a sentence backward? Or jumbling the words like a word salad? Take the sentence “The cat sat on the mat” as an example (Fig. 1). Imagine calculating how likely this exact sentence is to appear. Does it matter if you start with “The” and work forward, or begin with “mat” and go backward, or even shuffle the words randomly?...

When we are on the right side of an argument, most of us believe presenting the facts and supporting evidence should be enough to persuade others. Instead, we are baffled when friends and family continue to vote for policies that run counter to their interests or pour the milk before the tea. Presenting evidence is not enough for persuasion because people are motivated reasoners driven by their core values and...

Inspired by the brain, we find a goal-directed attention approach to feature reuse bests a commonly used machine learning strategy (Luo et al., 2020). In particular, attentional modulation of mid-level features in deep convolutional neural networks is more effective than retraining the last layer to transfer to a new task. Neuroscience and machine learning have been enjoying a virtuous cycle in which advances in one field spurs advances in...

What’s better than fMRI or cognitive modelling? Of course, their combination in the form of model-based fMRI! Rather than evaluating simple contrasts based on the experimental design, such as where in the brain lights up more for houses vs. faces, model-based fMRI evaluates proposed cognitive processes and representations. In this blog, we’ll first consider an example of how model-based fMRI reveals aspects of brain activity that would not easily...

When it comes to technology and communication, faster is usually considered better. For example, test pilot Chuck Yeager showed “The Right Stuff” by being the first person to break the sound barrier. We celebrate computer chips becoming faster. In the age of the internet, many people view real-time interactive communication, such as on Twitter (more on this later!), as highly desirable. However, faster is not always better. Resisting the...

My name is Sebastian Bobadilla-Suarez and I am an early career researcher (ECR — postdoc’ing in the Love Lab). I did my PhD with Brad Love at UCL as well. This post is about recent events regarding the review process of our manuscript titled Measures of neural similarity. Our manuscript was submitted to a prestigious journal and went through a formal review process. It was rejected by the reviewers,...

Imagine you think and work carefully on an ambitious paper for a few years, trying to answer fundamental questions the field has overlooked. Now, imagine after a very long wait you receive negative reviews that completely missed the main point. Instead, the reviews project a certain person’s pet concerns, goals, and interests onto your work, which are only tangentially related to your central questions. Worst yet, this person has...

Before Google, search engines, like AltaVista, often retrieved spurious web pages. Out of all the possible pages to return how does one determine which ones are the most relevant? One key to Google’s success was the PageRank algorithm developed by Google founders Sergey Brin and Larry Page in 1998. As they say, the rest is history, except there was a curious prehistory. Three years prior in 1995, while an...

Slogans can prove hollow or can invite one to reflect on core values. Aiming for the latter, our new lab motto is Inclusive, Productive, Accountable (IPA). We aim for a community where everyone is hoppy, I mean happy, no matter their drink of choice. In seriousness, for the IPA motto to have a positive effect, what it concretely means needs to be clear. Inclusivity This amounts to not forming...

Are we on the cusp of creating super-intelligent machines? Would such a super-intelligence put humanity at existential risk? Certainly, leaders in academia and industry are convinced that the danger of our own creations turning on us is real. For example, Elon Musk, founder of Tesla Motors and SpaceX, has set up a billion dollar non-profit company with contributions from tech titans, such as Amazon, to prevent an evil Artificial Intelligence...