Dear Friends and Students
The seed for automation celebrates 50 years!! Ted Hoff’s team from Intel
developed the 4004 a general-purpose processor that could be used across many
devices. The credit goes to Intel! Beyond, we need to appreciate another
company, requested Intel to develop this chip! That is Busicom, a Japanese
calculator maker (Nippon Calculating Machine Corp)!.
Busicom was a Japanese company that owned the rights to Intel's first
microprocessor, the Intel 4004, which they created in partnership with Intel in
1970. Had Busicom not allowed for general sale, probably things would have been
different, right! Busicom owned the exclusive rights to the design and its
components in 1970 but shared them with Intel in 1971!! This year we are
celebrating 50 years of this chip journey!!
I feel credit goes to both the founders of Busicom and Intel: Tadashi
and Robert!
Tadashi Sasaki was a
Japanese engineer, a founding member of Busicom, driving the development of the
Intel 4004 microprocessor, and later driving Sharp into the LCD calculator
market. Robert Norton Noyce, nicknamed "The Mayor of Silicon Valley",
was an American physicist who co-founded Fairchild Semiconductor in 1957 and
Intel Corporation in 1968.
While we are proud of
this legacy, recently Cerebras unveiled a chip that accelerates deep learning.
They developed Wafer-Scale Engine (WSE) – the largest chip ever built! It is 56x
larger than any other chip, the WSE delivers more compute, more memory, and
more communication bandwidth. The performance of a room full of servers into a
single unit the size of a dorm room mini-fridge!! What a miniature version?
Andrew is co-founder and
CEO of Cerebras Systems. Andrew holds a BA and an MBA from Stanford
University. Gary is co-founder and CTO of Cerebras Systems.
Gary
holds more than 50 patents. The perfect combination of Management and Technology!!
Let’s see the dimensions
of this chip: Sparse Linear Algebra Compute (SLAC) Cores 400,000, On-chip
Memory (SRAM), 18 GB SRAM, Memory Bandwidth 9.6 PB/sec, Interconnect Bandwidth100
Pb/sec, System I/O 1.2 Tb/s and Dimensions 15 rack units.
Dear Friends and Students
(Machine Learning Lovers)
The Cerebras software
platform integrated with popular machine learning frameworks like TensorFlow
and PyTorch. It also has a programmable C++ interface that allows researchers to
extend the platform and develop custom kernels.
I liked their tagline!! “Explore
More Ideas in Less Time. Reduce the Cost of Curiosity”. I am really amazed of
this statement- Reduce the Cost of Curiosity!
This year, many
engineering colleges in India started CSM and CSD courses (CSE- AL-ML) and (CSE-
DataScience). Cerebras Systems announced an Internship. Please read the following
requirements, which helps you to plan your AI-ML skills (cerebras.net/careers/).
The Role
Cerebras is developing
both novel algorithms to accelerate training of the existing neural network
architectures, as well as new, custom network architectures for the next
generation of deep learning accelerators. For this internship position, we are
looking for hands-on researchers who can:
Take an algorithm from
inception, to TensorFlow or PyTorch implementation, to results competitive with
state-of-the-art on benchmarks such as ImageNet classification.
- ·
Develop algorithms for training and
inference with sparse weights and sparse activations.
- ·
Develop algorithms for training at
unprecedented levels of scale and parallelism.
- · Publish results in Machine Learning
conferences and company messaging, like blog posts and white papers.
Skills &
Qualifications
- ·
Publications in Machine Learning such as
supervised, unsupervised, or reinforcement learning. Statistical modeling such
as generative modeling and probabilistic modeling.
- · Graduate and undergraduate students with a
background in Deep Learning and Neural Networks
- · Experience with deep learning models such
as Transformers, RNNs, and CNNs for language modeling, speech recognition, and
computer vision
- ·
Experience with high-performance machine
learning methods such as distributed training, parameter server, synchronous
and asynchronous model parallelism.
- ·
Experience with 16 bit / low precision
training and inference, using half-precision floating point, fixed point.
- ·
Experience with model compression and
model quantization.
Happy Deep Learning!!
Raise your Bar
Your Well-wisher
Ravi Saripalle
Join Inspire to Innovate
Storytelling Movement (i2itm.blogspot.com)
Source Inspiration:
www.bbc.com/news/technology-49395577
https://www.hindustantimes.com/photos/photos-how-the-microprocessor-changed-our-lives-for-you-101614411430009-8.html