张弛 @zhangchi_ai

Machine Learning Engineer Hong Kong Joined March 2021

Tweets

23
Followers

62
Following

254
Likes

94

Tom Zahavy @TZahavy

2 years ago

Great article by @StephenOrnes @QuantaMagazine on our recent AlphaZeroᵈᵇ paper: arxiv.org/abs/2308.09175!

Quanta Magazine @QuantaMagazine

2 years ago

Great article by @StephenOrnes @QuantaMagazine on our recent AlphaZeroᵈᵇ paper: arxiv.org/abs/2308.09175!

3 53 150 26K 29

1 6 18 2K 0

I'm super excited to share AlphaZeroᵈᵇ, a team of diverse #AlphaZero agents that collaborate to solve #Chess puzzles and demonstrate increased creativity. Check out our paper to learn more! arxiv.org/abs/2308.09175 A quick 🧵(1/n)

5 68 327 94K 109

Download Image

John Nay @johnjnay

3 years ago

ChatGPT for Training Data 1 ChatGPT rephrases each training sentence into multiple conceptually similar but semantically different sentences 2 Train smaller model Outperforms SoTA data augmentation methods for few-shot learning text classification Paper arxiv.org/abs/2302.13007

14 189 878 163K 373

Download Image

Ted Moskovitz @ted_moskovitz

3 years ago

Tired of ChatGPT screenshots? Miss the old days of watching RL agents walking around doing weird stuff? Look no further–I'm excited to share my @DeepMind internship project, where we develop a method to stabilize optimization in constrained RL. Link: arxiv.org/abs/2302.01275 🧵

9 59 697 120K 96

Download Gif

Tom Zahavy @TZahavy

3 years ago

Super proud to share ReLoad (arxiv.org/pdf/2302.01275…), an RL agent that converges in challenging constrained RL problems. The project was led by our fantastic intern @ted_moskovitz at the discovery team last summer. A thread 🧵

Ted Moskovitz @ted_moskovitz

3 years ago

9 59 697 120K 96

Download Gif

2 8 38 8K 4

Tom Zahavy @TZahavy

4 years ago

In our #Neurips2021 spotlight, we study RL problems where the goal is to minimize a cost over the state occupancy. When this cost is linear, we get the standard RL problem. When it is non-linear, we get apprenticeship learning, pure exploration, diversity and more. [1/7]

3 30 157 0 32

Download Image

Robert Lange @RobertTLange

4 years ago

How to overcome short-sightedness⌛️& ill-conditioned outer objectives in meta-learning❓BMG constructs a bootstrap target by updating the learner for a couple more steps 👣 Provides a new ATARI SOTA 🕹️ & opens up new perspectives 🎨#mlcollage [35/52] 📜: arxiv.org/abs/2109.04504

2 7 41 0 3

Download Image

Eric Xing @ericxing

4 years ago

My talk at the KDD Deep Learning Day on "It is time for DL to understand its expense bills", where I discussed how to build standardizable, adaptive, and automatable system for DL based on first principles underlying ML & Sys is now online. Enjoy watching youtube.com/watch?v=1ziZcH…

0 12 87 0 20

AK @_akhaliq

4 years ago

Reward is enough for convex MDPs pdf: arxiv.org/pdf/2106.00661… abs: arxiv.org/abs/2106.00661

0 16 55 0 9

Download Image

Markus Wulfmeier @m_wulfmeier

4 years ago

The diversity of emerged behaviour in nature is incredible. Now get your RL agent to discover this...

1 2 32 0 2

Download Video

Horace He @cHHillee

4 years ago

Haven't seen anybody else mention this, but Huawei just announced they trained a 200 BILLION transformer model - PanGu-α. This is bigger than GPT-3, but trained only for 40B tokens. Moreover, they're trained on an entirely Chinese stack: Huawei chips and Mindspore framework. 1/2

8 135 586 0 80

Download Image

AK @_akhaliq

5 years ago

Podracer architectures for scalable Reinforcement Learning pdf: arxiv.org/pdf/2104.06272… abs: arxiv.org/abs/2104.06272 "we argue that TPUs are particularly well suited for training RL agents in a scalable, efficient and reproducible way"

0 8 66 0 21

Download Image

Machine Learning Street Talk @MLStreetTalk

5 years ago

We go into full inception mode with @DeepMind research scientist Dr. Tom Zahavy @TZahavy and discuss metagradients in reinforcement learning! With @ykilcher , @RobertTLange and @ecsquendor

2 6 48 0 3

Download Image

Lilian Weng @lilianweng

5 years ago

Just finished the first episode. Very high quality interview and strongly recommend it :) Now stepping into the second episode woohoo

Pieter Abbeel @pabbeel

5 years ago

Just finished the first episode. Very high quality interview and strongly recommend it :) Now stepping into the second episode woohoo

4 36 199 0 17

3 11 133 0 13

Hugging Face @huggingface

5 years ago

Last week, EleutherAI released two checkpoints for GPT Neo, an *Open Source* replication of OpenAI's GPT-3 These checkpoints, of sizes 1.3B and 2.7B are now available in🤗Transformers! The generation capabilities are truly🤯, try it now on the Hub: huggingface.co/EleutherAI/gpt…

17 302 1K 0 149

Download Image

hardmaru @hardmaru

5 years ago

Jeff Bezos ✕ Elon Musk ＝ Dr. Evil²

0 5 100 0 1

Yann LeCun @ylecun

5 years ago

Madgrad: not an angry graduate student, but a new stochastic optimization method for Deep Learning.

AI at Meta @AIatMeta

5 years ago

Madgrad: not an angry graduate student, but a new stochastic optimization method for Deep Learning.

23 491 2K 0 203

Download Image

16 164 939 0 57

/MachineLearning @slashML

5 years ago

Generic template to bootstrap your PyTorch project with PyTorch Lightning, Hydra, W&B, DVC, and Streamlit reddit.com/r/MachineLearn…

1 16 63 0 13

Kevin K. Yang 楊凱筌 @KevinKaichuang

5 years ago

Instead of fine-tuning an entire neural network, use its Jacobian to define a Gaussian process prior and then compute the closed-form GP posterior on new data. Wesley J. Maddox @Shuai93Tang @andrewgwils @pgmoren @adamianou arxiv.org/abs/2103.01439…

6 93 474 0 129

Download Image

Ilya Kostrikov @ikostrikov

5 years ago

I’m excited to release my implementations of RL algorithms in Jax: github.com/ikostrikov/jax…. On my machine with Titan X (Pascal), it works more than twice faster (4x faster in some cases) than similar PyTorch implementations. Thanks to @nikishin_evg for helping with Jax!