Why should economic variables such as reward and effort affect movement vigor?
We developed a new paradigm to study the link between decision-making and vigor.
biorxiv.org/content/10.110
Follow
Elahe Arani
@ElaheArani
Senior AI manager & senior research scientist
Incoming assistant professor
Working at the intersection of #neuroscience and #AI
Eindhoven, NederlandJoined June 2022
Elahe Arani’s Tweets
وعده دیدار ما بیستم فوریه ساعت ۱۱:۳۰
، بروکسل
💚گام دوم
🤍با عزم راسخ
❤️تا پیروزی
حضور باشکوه و همبسته ما پیام اراده ملت ایران را به شایستگی به شورای اروپا میرساند؛
قطعنامه تروریستی دانستن #IRGCterrorists که به تصویب پارلمان🇪🇺 رسید را اجرایی کنید.
#BRUSSELS20FEB
732
7,775
17.3K
We show that using MIM not only improves the performance of monocular depth estimation but also results in #robustness against noise, blur, weather conditions, digital artifacts, occlusions, and untargeted and targeted adversarial attacks. 2/2
#neuralnetworks #deeplearning #ML
1
Show this thread
Our paper, "Image Masking for Robust Self-Supervised Monocular Depth Estimation" has been accepted at #ICRA2023! We adapt masked image modeling for direct training to improve self-supervised monocular depth estimation.1/2
arxiv.org/pdf/2210.02357
#DL #AI #computervision #research
2
1
14
Show this thread
Yoshua Bengio, Joëlle Pineau, Foutse Khomh and Catherine Régis of Mila discussed ChatGPT and AI more broadly with Alexandre Sirois of .
Read the article here ➡️ pulse.ly/bh5db9a9kl (french only)
7
13
Our findings suggest that a principled mechanism for learning from low errors in a dual-memory system provides a promising approach to mitigating abrupt representation drift in CL that warrants further study.
2
Show this thread
ESMER further enables learning under high levels of label noise, crucial for real-world applications. 4/5
1
1
Show this thread
By leveraging error history and utilizing Error-Sensitive Reservoir Sampling, ESMER effectively reduces abrupt drift in representations and task-recency bias in CL. 3/5
1
1
Show this thread
Unlike traditional deep neural networks, the proposed approach decreases the sensitivity to errors as a function of their magnitude, similar to the learning mechanism in the human brain. 2/5
#neuroscience #AI #MachineLearning
1
2
2
Show this thread
Our paper Error Sensitivity Modulation based Experience Replay is accepted at #iclr2023. We tackle the challenge of catastrophic forgetting in #ContinualLearning by modulating the error sensitivity in a dual-memory rehearsal-based system. 1/5
openreview.net/pdf?id=zlbci70
#DeepLearning
1
10
32
Show this thread
Considering the current situation in #Iran, it is crucial that he be helped in obtaining a visa and leaving the country without delay. 3/3
Show this thread
Despite my efforts, I am unable to offer a #PhDposition to him/her due to a sudden funding issue. I know that he would be an asset to any organization in the #AI field and I am deeply disappointed that I am unable to help him at this time. 2/3
1
Show this thread
Sharing an exceptional #AI talent from Iran, top 20 out of 500k students, seeking #Research or #Engineering role in industry or #PhD in #ContinualLearning. Ideal candidate for top institutions. Contact me for details. 1/3
#funding #opportunities #AI #ML #talent #job
1
1
6
Show this thread
Our aim is to inspire both the #research and #industrial communities towards a new direction in network design and evaluation, with a focus on a comprehensive and far-reaching impact. Keep an eye out for an upcoming video presentation. 2/2
2
Show this thread
From accuracy to energy efficiency and robustness, our #tmlr (Survey Certification) paper serves as a handbook for both #academia and #industry with a focus on the modern #RealTimeDetection. 1/2
openreview.net/pdf?id=ywr5sWq
#ObjectDetection #ComputerVision #Handbook #deeplearning
3
10
31
Show this thread
Our paper, "Task-Aware Information Routing from Common Representation Space in Lifelong Learning," has been accepted at the #ICLR2023!
Our proposed CL method, TAMiL, is inspired by the Global Workspace Theory of conscious information access in the brain. 1/4
#ContinualLearning
2
6
26
Show this thread
LURE exhibits robustness in solving more common challenges in real-world problems, including
learning with noisy labels, natural corruption, and adversarial attacks. 4/4
2
Show this thread
We show that LURE provides consistent performance gains across datasets in both classification and few-shot settings. 3/4
1
1
2
Show this thread
LURE interchanges between the unlearning
phase, which selectively forgets the undesirable information in the model through weight
reinitialization in a data-dependent manner, and the relearning phase, which emphasizes
learning on generalizable features.2/4
#ArtificialIntelligence
1
1
3
Show this thread
Our paper "Learn, Unlearn & Relearn: An Online Learning Paradigm for Deep Neural Networks" has been accepted at #TMLR. Our work presents a novel online learning paradigm, LURE, that balances efficiency and generalization.1/4
openreview.net/forum?id=WN1O2)
#onlinelearning #deeplearning
4
36
198
Show this thread
Quote Tweet
You don't need a #PhD or to work at a top company or lab to succeed in AI research. Our team publishes papers on neuro-inspired AI with researchers holding just bachelor/master's degrees. Stay tuned for more info.
#AI #MachineLearning #DataScience
2
5
Our experimental results show that TAMiL outperforms current state-of-the-art rehearsal- and regularization-based methods on existing benchmarks, bridging the gap between fixed capacity and parameter isolation approaches while being scalable. 4/4
Show this thread
We employed simple, undercomplete autoencoders to create a communication bottleneck between the common representation space and the global workspace, allowing only the task-relevant information to the global workspace. 3/4
#deeplearning #MachineLearning
1
1
Show this thread
It utilizes task-attention modules to capture task-specific information from a common representation space, effectively reducing task interference and mitigating catastrophic forgetting. 2/4
1
2
Show this thread
Our paper, "Task-Aware Information Routing from Common Representation Space in Lifelong Learning," has been accepted at the #ICLR2023!
Our proposed CL method, TAMiL, is inspired by the Global Workspace Theory of conscious information access in the brain. 1/4
#ContinualLearning
2
6
26
Show this thread
Summary Of AC-reviewer Meeting: At the video conference, the reviewers were all in complete agreement that the method was novel, but also that the empirical evaluation was unconvincing.
Show this thread
This raises concerns about why being novel and well-written is NOT enough for a research paper to be accepted for publication. It also brings to light the need for a more robust evaluation process that does not rely on tips and tricks to outperform existing methods. (2)
1
3
Show this thread
In addition to 2 accepted papers at #ICLR2023, we have a rejected paper. The method presented in the paper was deemed novel by all reviewers, yet the empirical evaluation was found unconvincing. (1)
1
6
Show this thread
Our two papers have been accepted at #ICLR2023. We are grateful for the hard work of the reviewers. Let's continue to strive for excellence in the review process and explore ways to engage reviewers in discussion sessions for even more robust evaluations.
#ContinualLearning #AI
2
3
29
Show this thread
Check it out to learn how our method learns modular and specialized representations while maintaining reusability.
Coauthors:
#MachineLearning #ArtificialIntelligence
2
Show this thread
Our paper "Dynamically Modular and Sparse General Continual Learning" has been accepted by VISAPP-2023. We introduce a new method to prevent catastrophic forgetting in DNNs by incorporating dynamic modularity and sparsity.
#deeplearning #ContinualLearning
arxiv.org/pdf/2301.00620
2
1
7
Show this thread
Our request is clear: put the Islamic Revolutionary Guard Corps (IRGC) on the terrorist list.
#IRGCterrorists
#WomanLifeFreedom
#MahsaAmini
3,306
20.4K
42.3K
💯more than 100 French senators 🇫🇷are fighting against #execution #StopExecutionsInIran #IranRevolution2023
302
601
1,267
We propose a novel approach to mimic human-like efficient continual learning in DNNs, using sparse coding and multiple memory replay to balance reusability and interference of features across tasks.
Coauthors:
3
Show this thread
Our paper, "Sparse Coding in a Dual Memory System for Lifelong Learning," has been accepted as an Oral presentation at #AAAI23.
#deeplearning #artificialintelligence #MachineLearning
1
1
28
Show this thread
#ArshiyaTakdastaan, 18, is charged with "corruption on earth", "waging war against God", "seduction and "incitement of people to war and killing" under forced confession. He has been sentenced to death. Please find him a political sponsor and be his voice!
#IranRevolution
4
77
37
#JavadRouhi, 35, from Amol who was abducted on September 22 by the regime forces has been sentenced to death in Sari revolutionary court for the crimes of "waging war against God" and "corruption on earth". He's been forced to confess against his own interest.
#IranRevolution
3
78
40
#FarhadTahazadeh & #FarzadTahazadeh,Farzad, 25, and his brother Farhad, 26, were arrested in #Oshnavie on Sep 25th, accused of "waging war against God", were sentenced to death in their absence in the court session. Their lives in a serious danger. Be his voice!
#IranRevolution
18
150
75