A bit about me

Hi! My name is Jen and I am a Ph.D. student in machine learning at IST Austria, advised by Dr. Dan Alistarh. I was also fortunate to intern with Weiwei Yang and Janardhan Kulkani at Microsoft Research.

I am primarily interested in the practical and theoretical foundations of applying machine learning and deep learning models in real-life–specifically, questions of efficiency, fairness, bias, and general model interpretability. I study these specifically in the context of edge-device deployment and other constrained-resource environments.

Before starting at IST, I spent some time in industry: I traded equity options on Wall Street, taught math and science to high-school students in San Francisco, helped build a healthcare startup, and built and deployed deep learning models for text recognition at Google LA.

I am a community lead at Cohere for AI, where I host talks on using ML for social good.

Publications

SPADE: Sparsity-Guided Debugging for Deep Neural Networks.

Arshia Soltani Moakhar, Eugenia Iofinova*, Elias Frantar, Dan Alistarh

ICML 2024, 2024-07-21

We improve DNN interpretability by computing a sparse trace of an input through a model prior to running interpretability methods.

Download here

Panza: A Personalized Text Writing Assistant via Data Playback and Local Fine-Tuning

Armand Nicolicioiu, Eugenia Iofinova*, Eldar Kurtic, Mahdi Nikdan, Andrei Panferov, Ilia Markov, Nir Shavit, Dan Alistarh

ArXiv, 2024-06-24

We demonstrate the feasibility of training an e-mail composition assistant entirely on a consumer-grade GPU.

Download here

Accurate Neural Network Pruning Requires Rethinking Sparse Optimization

Denis Kuznedelev*, Eldar Kurtic*, Eugenia Iofinova*, Elias Frantar, Alexandra Peste*, Dan Alistarh

TMLR, 2024-06-20

We show that, generally speaking, dense training settings are not optimal for sparse training for the same dataset/architecture.

Download here

SparseProp: Efficient Sparse Backpropagation for Faster Training of Neural Networks

Mahdi Nikdan*, Tommasso Pegolotti*, Eugenia Iofinova, Eldar Kurtic, Dan Alistarh

ICML 2023, 2023-07-23

My groupmates found a way to do faster backpropagation through unstructured-sparsity weights.

Download here

Bias in Pruned Vision Models: In-Depth Analysis and Countermeasures

Eugenia Iofinova, Alexandra Peste, Dan Alistarh

CVPR 2023, 2023-06-18

We demonstrate that ‘stereotyping’, i.e., amplifying feature correlations, increases with model sparsity, thus leading to increased bias.

Download here

How Well Do Sparse ImageNet Models Transfer?

Eugenia Iofinova*, Alexandra Peste*, Mark Kurtz, Dan Alistarh

CVPR 2022, 2022-06-20

We investigate the effect of the pruning method on transfer learning.

Download here

FLEA: Provably Fair Multisource Learning from Unreliable Training Data

Eugenia Iofinova*, Nikola Konstantinov, Christoph H. Lampert

Preprint, 2021-06-22

In which we propose a theoretically rigorous algorithm for detecting and eliminating malignant (perturbed) data in a multisource setup.

Download here

AC/DC: Alternating Compressed/DeCompressed Training of Deep Neural Networks

Alexandra Peste, Eugenia Iofinova, Adrian Vladu, Dan Alistarh

NeurIPS 2021, 2021-05-21

In which we refine the principle of Iterative Hard Thresholding to propose a simple and effective protocol for unstructured sparse training.

Download here

The VRNetzer platform enables interactive network analysis in Virtual Reality

Sebastian Pirch, Felix Müller, Eugenia Iofinova, Julia Pazmandi, Christiane V. R. Hütter, Martin Chiettini, Celine Sin, Kaan Boztug, Iana Podkosova, Hannes Kaufmann, Jörg Menche

Nature Communications, 2021-01-19

3D interactive platform to explore biological networks

Download here

CV

Download my CV here.