Ayoub Ajarra
@AjarraAyoubPhD student at Scool (Previously Sequel) - Inria Lille
Similar User
@AhrenJin
@hyanan16
@xuefeng_du
@XtremSup
@ShuruiGui
@YizhouWang14
@XinmingHou
@dmarviss
@KhalidDinar4
@lozanorcintaes1
@KruskalLin
@chunhuizng
When a measure becomes a target it ceases to be a good measure This adage from Marilyn Strathern known as Goodhart's law, has implications beyond economics, public policy & AI With @le_science4all we tried few years ago to formalise it, here it is finally arxiv.org/abs/2410.09638
Is "offline RL" in offline-to-online RL really necessary? Surprisingly, we find that replacing offline RL with *unsupervised* offline RL often leads to better online fine-tuning performance -- even for the *same* task! Paper: arxiv.org/abs/2408.14785 🧵↓
I'm going to write a survey on bounding the total variation distance, focusing on the case of product measures send me your favorite inequalities PS I know all of these 👇
I don't know how someone came up with this crazy formula for the mean of a logit-Normal, but I'm glad they did. It converges extremely fast. en.wikipedia.org/wiki/Logit-nor…
Excited to share that our work with Karolina, Mahdi, Roi, and Dan has been selected for a best paper award 🎉
hree useful identities for min(u,v) each one gets used in my upcoming paper stay tuned
Temporal distances (expected number of time steps between states) in stochastic MDPs in general lack metric structure. It has been a long-standing question how to design a (quasi)metric notion of "distance" in such settings. In a new paper, we have a viable solution! 🧵👇
🧵New paper: Machine Unlearning Fails to Remove Data Poisoning Attacks, ft @MartinPawelczyk, @jimmy_di98, @ayush_sekhari, @SethInternet Title says it all: current approaches for machine unlearning (MUL) are not effective at removing the effect of data poisoning attacks. 1/n
We've derived tight lower and upper bounds for differentially private finite-armed & linear bandits, while we lack the same for contextual bandits. At #COLT2024, @achraf_azize presents open problems in contextual bandits with privacy. @BasuDebabrota @Inria_Lille @RechercheUlille
#phdlife ✌️Dans les algorithmes Top Two, un choix considéré "leader" est opposé à un "challenger" pour choisir la meilleure option. 📃Dans sa #thèse, @MarcJourdan5 de l’équipe @InriaScool en fait une méthode offrant garanties théoriques et excellentes performances empiriques.
[📜preprint] New results for an old problem We revisit the basic statistics problem "simple binary hypothesis testing": Given (i) two distributions p & q and (ii) n data points promised to be sampled either i.i.d. from p or i.i.d. from q, identify the true distribution
AI researcher Minqi Jiang says the next frontier of AI is moving from to systems that answer questions to systems which ask the questions
🧵New paper: "Not All Learnable Distribution Classes are Privately Learnable" to appear in #ALT2024. We refute a conjecture of @ashtiani_hassan We show that there exists a learnable distribution class which is not privately learnable. w @markmbun @argymouz @vkerdos 1/n
As semester draws to end, I want to share this *identity* (h/t @tengyangx) that connects so many fundamental pieces of the RL theory together: optimism, pessimism, policy opt, proved by PD lemma + Bellman-error telescoping, all in one equation! 1/3
I hope Q* does not involve RL because I was hoping I will actually never have to learn that god-awful RL notation. Please let it be anything else but RL, please.
"Toward General Virtual Agents" I recently gave a talk at MIT. I argued that we should use tools from reinforcement learning and search to improve the capability and alignment of LLM agents. Slides: drive.google.com/file/d/1kDvmrm… Video:
United States Trends
- 1. Travis Hunter 7.302 posts
- 2. $CUTO 8.099 posts
- 3. Northwestern 5.629 posts
- 4. Sheppard 2.911 posts
- 5. Carnell Tate 1.039 posts
- 6. Colorado 66,2 B posts
- 7. Arkansas 26,9 B posts
- 8. Denzel Burke N/A
- 9. Ewers 1.123 posts
- 10. Shedeur 2.880 posts
- 11. $CATEX N/A
- 12. Jahdae Barron N/A
- 13. Wrigley 3.517 posts
- 14. Jeremiah Smith N/A
- 15. #SkoBuffs 2.982 posts
- 16. #HookEm 2.321 posts
- 17. #collegegameday 5.546 posts
- 18. #Buckeyes N/A
- 19. Renji 6.774 posts
- 20. Gus Johnson N/A
Who to follow
-
Yiqiao Jin @EMNLP2024
@AhrenJin -
Xiner Li
@hyanan16 -
Sean Xuefeng Du (on academic job market)
@xuefeng_du -
Xiusi Chen
@XtremSup -
Shurui Gui
@ShuruiGui -
Yizhou Wang
@YizhouWang14 -
Dylan X. Hou
@XinmingHou -
Marvis
@dmarviss -
Khalid Dinar
@KhalidDinar4 -
clr
@lozanorcintaes1 -
Yuchao Lin
@KruskalLin -
Chunhui Zhang
@chunhuizng
Something went wrong.
Something went wrong.