Scaling Up Test-Time Compute with Latent Reasoning with Jonas Geiping - #723

Scaling Up Test-Time Compute with Latent Reasoning with Jonas Geiping - #723

Today, we're joined by Jonas Geiping, research group leader at Ellis Institute and the Max Planck Institute for Intelligent Systems to discuss his recent paper, “Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach.” This paper proposes a novel language model architecture which uses recurrent depth to enable “thinking in latent space.” We dig into “internal reasoning” versus “verbalized reasoning”—analogous to non-verbalized and verbalized thinking in humans, and discuss how the model searches in latent space to predict the next token and dynamically allocates more compute based on token difficulty. We also explore how the recurrent depth architecture simplifies LLMs, the parallels to diffusion models, the model's performance on reasoning tasks, the challenges of comparing models with varying compute budgets, and architectural advantages such as zero-shot adaptive exits and natural speculative decoding. The complete show notes for this episode can be found at https://twimlai.com/go/723.

Episoder(778)

Milestones in Neural Natural Language Processing with Sebastian Ruder - TWiML Talk #195

Milestones in Neural Natural Language Processing with Sebastian Ruder - TWiML Talk #195

In this episode, we’re joined by Sebastian Ruder, PhD student studying NLP at National University of Ireland and Research Scientist at text analysis startup Aylien. We discuss recent milestones in neu...

29 Okt 20181h 1min

Natural Language Processing at StockTwits with Garrett Hoffman - TWiML Talk #194

Natural Language Processing at StockTwits with Garrett Hoffman - TWiML Talk #194

In this episode, we’re joined by Garrett Hoffman, Director of Data Science at Stocktwits. Stocktwits is a social network for the investing community which has its roots in the use of the $cashtag on T...

25 Okt 201850min

Advanced Reinforcement Learning & Data Science for Social Impact with Vukosi Marivate - TWiML Talk #193

Advanced Reinforcement Learning & Data Science for Social Impact with Vukosi Marivate - TWiML Talk #193

In the final episode of our Deep Learning Indaba series, we speak with Vukosi Marivate, Chair of Data Science at the University of Pretoria and a co-organizer of the Indaba. My conversation with Vuko...

23 Okt 201846min

AI Ethics, Strategic Decisioning and Game Theory with Osonde Osoba - TWiML Talk #192

AI Ethics, Strategic Decisioning and Game Theory with Osonde Osoba - TWiML Talk #192

In this episode of our Deep Learning Indaba Series, we’re joined by Osonde Osoba, Engineer at RAND Corporation. Osonde and I spoke on the heels of the Indaba, where he presented on AI Ethics and Poli...

18 Okt 201847min

Acoustic Word Embeddings for Low Resource Speech Processing with Herman Kamper - TWiML Talk #191

Acoustic Word Embeddings for Low Resource Speech Processing with Herman Kamper - TWiML Talk #191

In this episode of our Deep Learning Indaba Series, we’re joined by Herman Kamper, lecturer at Stellenbosch University in SA and a co-organizer of the Indaba. We discuss his work on limited- and zero...

16 Okt 20181h 1min

Learning Representations for Visual Search with Naila Murray - TWiML Talk #190

Learning Representations for Visual Search with Naila Murray - TWiML Talk #190

In this episode of our Deep Learning Indaba series, we’re joined by Naila Murray, Senior Research Scientist and Group Lead in the computer vision group at Naver Labs Europe. Naila presented at the In...

12 Okt 201841min

Evaluating Model Explainability Methods with Sara Hooker - TWiML Talk #189

Evaluating Model Explainability Methods with Sara Hooker - TWiML Talk #189

In this, the first episode of the Deep Learning Indaba series, we’re joined by Sara Hooker, AI Resident at Google Brain. I spoke with Sara in the run-up to the Indaba about her work on interpretabilit...

10 Okt 20181h 3min

Graph Analytic Systems with Zachary Hanif - TWiML Talk #188

Graph Analytic Systems with Zachary Hanif - TWiML Talk #188

In this, the final episode of our Strata Data Conference series, we’re joined by Zachary Hanif, Director of Machine Learning at Capital One’s Center for Machine Learning. We start our discussion wit...

8 Okt 201854min

Populært innen Politikk og nyheter

giver-og-gjengen-vg
aftenpodden
aftenpodden-usa
forklart
stopp-verden
popradet
det-store-bildet
fotballpodden-2
dine-penger-pengeradet
rss-gukild-johaug
bt-dokumentar-2
nokon-ma-ga
lydartikler-fra-aftenposten
aftenbla-bla
hanna-de-heldige
rss-dannet-uten-piano
e24-podden
frokostshowet-pa-p5
rss-ness
rss-penger-polser-og-politikk