Poornima Ramesh

Data Scientist at IDinsight

[The Agency Fund @AgencyFund]

🔍 How can social sector leaders access decision-relevant data faster?

One way is with Ask-A-Metric (AAM), a tool built by @IDinsight to help reduce technical barriers to data access and use.

Join us with co-hosts Project Tech4Dev and @KabakooLabs for a dynamic #LLMsInSocialSector webinar, where IDinsight will discuss the technology behind AAM and share learnings on building AI-powered tools.

📅 Thursday, June 13, 2024
🎦 Zoom

Register for the event so you don’t miss this: https://t.co/PSnVqCtysY

#GenAI #SocialImpact

[IDinsight @IDinsight]

#AI is increasingly being relied upon by decision-makers because the technology has proven efficiencies. But for decisions to be unbiased, accurate and equitable, the data used to train ML models must be high-quality and representative.

As our teams blaze the trail for AI use in the social sector, we are cognizant of these challenges and are committed to creating and maintaining equitable machine learning models. 🔗 Learn more: https://t.co/iT586XwfQY

Interested in using machine learning to amplify your organisation's impact? IDinsight can support the entire life cycle of model development and deployment - learn about our work and contact our team: https://t.co/dTG6BXEUS2

[IDinsight @IDinsight]

#Top5blogs NUMBER 4️⃣: You’ve planned your research design, assigned schools to treatment and control groups, conducted data collection, and are ready to dive in and clean your data set -- only to realize your research design HASN’T been followed! What do you do? Start over? Or can you still learn something useful about the intervention?

In their blog, our team Jeff (@jefferyjmcmanus), Michael (@MichaelSebele), Mtise shares how they dealt with such a situation and generated meaningful insights despite having their #RCT design upended!

🔗 Read the blog here: https://t.co/pmuCxs5wQZ

[IDinsight @IDinsight]

🥁 As we kick off 2024, we are sharing our 5 most-read #blogs from last year! 🥁

NUMBER 5️⃣ ~ Satellite imagery combined with AI can help make important predictions for #globaldev practitioners, for example, of forest cover or population census information. #MOSAIKS, created by Esther Rolf (@rolf_comma_e) and colleagues, allows any user to enter GPS coordinates and extract satellite imagery information. In this blog, IDinsight's Data Science and Engineering Team shares how they extended the tool (and open-sourced the package) to speed it up, add features for #Kenya and #India, and look at specific time periods and/or resolutions. #AIforGood

🔗 Read the blog: https://t.co/zBNcRmr6JG

🔗 Delve into our 2023 Year in Review and read about the impactful work done by our Data Science and Engineering Team: https://t.co/nOLWoWSsuN

@pramesh95 @amiiremami @sid_ravinutala

[Tim Vogels @TPVogels]

Over the years my lab has been working on #meta_learning #plasticity rules in #spiking networks. Here's progress report on how far we can get using a twist on simulation based inference (fSBI), presented at @NeurIPSConf (#405) w/ @basile_cfx, @pramesh95 @ppjgoncalves & @jakhmack.

[Jakob Macke @jakhmack]

Just quick: Hugely excited to be given the chance to work on a new project (more details soon) on building network models of large-scale neural computations. Many thanks to the fantastic @mackelab @ml4science, @srinituraga + @lappalainenjk for the joint work on which this builds!

[sbi developers @sbi_devs]

🔔It’s Neural Ratio Estimation (NRE) time🎄! We are happy to announce that two new NRE methods have been contributed to the SBI toolkit: Many thanks to
@bkmi13 for contributing Contrastive NRE, and to
@ArnaudDelaunoy for contributing Balanced NRE. 🕯️/ 4

[Jonas Beck @__jnsbck__]

Stoked to share my first #NeurIPS2022 paper.

"Efficient Identification of informative features in SBI"

https://t.co/Z7GrcuJsDK

w. @deismic_, @yvesbernaerts, @jakhmack, @CellTypist
@ml4science

We use NLE to efficiently identify how features constrain posteriors in SBI. (1/8)

[ML⇌Science Colaboratory @mlcolab]

Our preprint "Spatiotemporal modeling of European paleoclimate using doubly sparse Gaussian processes" is on @arxiv! This is one of the outcomes of a cooperation we (@sethaxen @alpiges @alvorithm) are currently running with @sommer_geo and Nils Weitzel!

https://t.co/U98xCHHSbo

[sbi developers @sbi_devs]

📢 We just released a new sbi version with many improvements and several new features: a SNPE-method that avoids “leakage”, expected coverage tests, and embedding networks that allow iid data for SNPE 🧵 1/6

[Machine Learning in Science @mackelab]

Can circuits generate functional, energy-efficient, and temperature-robust neural activity with widely disparate sets of membrane and synaptic conductances? Check it out in our latest work, now out @PNASNews https://t.co/nKdG6lTv10! By @deismic_ @jakhmack @ppjgoncalves

[Jakob Macke @jakhmack]

Now in final form, many congrats to @deismic_ , and to @ppjgoncalves for his first last-author paper! This paper benefited greatly from peer review: it was first rejected post peer review from @elife — sure that hurt but the reasons were understandable and highly constructive.

[Tim Vogels @TPVogels]

Dear #neurotwitter, we dropped this onto @biorxiv_neursci for your perusal. It's an unusual piece from my lab & I really, really enjoyed working on it with @chc1987. Here it is: Metabolically spikes serve neuronal energy homeostasis (and protect neurons). https://t.co/muKt2jNPyG

[IDinsight @IDinsight]

(2/2) Bring your skills to the test and try to classify crops across agricultural fields in Northern India! Know someone who’d be interested in the challenge? Tag them in the comments!
🗓Competition closes on October, 31. https://t.co/i4Usv99T5K

[Machine Learning in Science @mackelab]

it's got human single units, theta oscillations, phase coding, and RNNs - what more could you want???

[Jakob Macke @jakhmack]

Preprint: @steliebe J Niediek et al @humansingleunit recorded 100s of single units in humans, found funky firing during memory tasks. @matthijs_pals task-trained RNNs which exhibit similar patterns and point to potential mechanisms. (Also, cameos by the 👸 and Stable Diffusion.)

[Stefanie Liebe @steliebe]

I don’t remember what life was like before starting this project years ago, but somehow our brains track the order of events well. How? See our paper on spike phase coding in human MTL and RNNs with J.Niediek @matthijs_pals @mackelab @humansingleunit https://t.co/AQJLdNanEx 1/5

[Jakob Macke @jakhmack]

Wondering how wiggly RNNs do wiggly computations? Dont miss @matthijs_pals talk at @neuromatch #nmc22 on -- joint work with @steliebe @_rdgao and Omri Barak!

[Machine Learning in Science @mackelab]

The brain can encode information in the phase of oscillations. Want to find out how recurrent neural networks, popular models of neural processing, do this? Watch a short recorded talk by @matthijs_pals at #nmc22 https://t.co/EweWdpPRPH
-Co-supervised by Omri Barak

[Poornima Ramesh @pramesh95]

Elegant and blazingly fast work (from the authors, and the method too)!

[Machine Learning in Science @mackelab]

Simulation-based inference (SBI) has accelerated model-building across many disciplines. It’s also useful for studying cognition: We present Mixed Neural Likelihood Estimation, an SBI method tailored to models of decision-making! By @janfiete @janmatthis @_rdgao @jakhmack
1/5