The way we remember people

Posted by : at

Category : Personal


I got to meet my parents and relatives after 3 years, on a tiring road trip across the West Coast (Santa Barbara – Las Vegas – Mammoth Lakes – Yosemite – San Francisco – Redwood Forest – Portland, about 2280 miles in 6 days).

More than the memories we made throughout the trip, I was struck by a lot of things. Being in AI research for some time now, memory is something that keeps me super curious. Typically, memory for agents can be thought of as being of two kinds: episodic and internal. This might not be the exact terminology (that’s why this isn’t a tech blog), but this is how I keep it in mind.

Episodic is where you explicitly keep track of events; it’s like taking a video of the event. You can replay the video to exactly verify the list of things that happened. Meanwhile, Internal is where you, in some way, remember the event because you’ve experienced it. You have a trace of it in your mind, but you might not be able to verify things by replaying the memory. You might be able to visualize things, but it’s hard to guarantee that you remember everything the exact way it happened.

In my head, I believed I had been changing over time. I believed I’d overcome most of what appeared to be problems or things people had pointed out as my problems. But while I met with my parents, I realized that I haven’t changed much. “-Some things, they never change-“ (Passenger – Everything). I thought I had become more of a patient person, but turns out, I have not. Maybe the function of getting annoyed is not just dependent on the person themselves, but also on the context and much more minute things. I thought I was not annoyed by a certain event X anymore, but turns out, annoyance depends on who does a certain event too.

Now that’s very scary. What this told me was that the issues I’ve worked on aren’t gone yet. Maybe from a meta-learning perspective, I should be working on the same event X but with multiple people and maybe then it will generalize better to a wider range of people I haven’t worked with?

Another interesting aspect was that I realized the way I remembered my parents also surprised me. Most of these things are a form of Internal memory? So whatever you remember is in the weights? (Neurons of the brain?) For some reason, I felt that my memory of people was very different, but when I met some relatives in person, my interactions with them reminded me of who they are (not necessarily good or bad). Then I realized, it’s no different from how I remember myself. ‘Self’ is a very funny concept, haha. This is kinda similar to the ‘hallucination problem’ in AI. The main difference I see is: in AI, these hallucinations don’t change with time unless I change the weights myself.

Meanwhile, I feel like, with time, the internal memory starts getting modulated? Maybe it’s an evolutionary trait? Memories and events that one remembers of people start slightly getting diluted toward the good side? This includes oneself too! I thought I had handled things better, and so did my parents and people I know, but when I replay it with ‘truer’ facts, I realize that wasn’t the truth.

Maybe, naturally, our brains alter the ‘internal memory’ of ourselves to be considered good/nice people, to prolong our life? That would explain why people are very self-centered—sometimes justifying every action they make. Well, in that case, it might not necessarily be just people around them supporting them through everything that spoils them and makes them think they’re always right. But the immediate counter I can think of is how people around me hold anger/hatred toward people who are not around them and hold it pretty strong! Maybe it’s a function of time, and it might change? (Maybe not linearly, that they might hate but then start letting the hate go?) Would love to have someone explain this clearly to me.

About Vihaan Akshaay

I am an Applied AI Researcher with first-author publications at top-tier venues, including ICLR 2025 and NeurIPS 2023, in Computer Vision and Deep Reinforcement Learning. My work spans five research internships across premier institutions, including The Jackson Laboratory (JAX), IIT Madras, Georgia Tech, NTU Singapore, and a joint role at UC Santa Barbara and Carnegie Mellon University.

My research bridges disciplines—developing AI systems for biological behavior analysis, robotics, mechanical systems, and Earth sciences. At IIT Madras, I led the iBot Robotics Club and co-developed the ARTEMIS Railroad Crack Detection Robot, winning the International James Dyson Award. My Master’s thesis on unsupervised behavior recognition in mice was advised by B. Ravindran and Dr. Vivek Kumar at JAX.

I recently completed my M.S. in Computer Science at UC Santa Barbara, working under Lei Li and Yu-Xiang Wang. Inspired by human problem-solving strategies, I proposed a bi-directional framework for goal conditioning in state-space search. I also introduced an edge-attention-based U-Net for environmental segmentation and helped curate a large-scale landslide detection dataset with Gen Li using 40 years of Landsat imagery.

Other projects include analyzing the stability of Deep Q-Networks with Siva Theja Maguluri at Georgia Tech and designing kernelized deep randomized models (eDRVFLs) with P. N. Suganthan at NTU Singapore.

I specialize in translating cutting-edge AI theory into practical, high-impact solutions across domains. I am currently seeking opportunities in applied AI research or machine learning engineering roles, particularly those focused on impactful, real-world applications.

Useful Links