bug/partition_pdf incorrectly parsed math formula from table

Describe the bug
I'm working on RAG project to assist in learning AI papers. I used unstructured to parse texts and tables from pdf papers (I used Attention Is All You Need paper in this case). When I checked the parsed results of tables, I found several cases of incorrectly parsed 'power of' and 'dot multiplication symbol',
For example this time complexity O(n? - d), it supposed to return O(n2 . d) or better yet O(n^2 . d) (if it's possible).
Either this partition_pdf bug
or
Is there any other method that I need to try to get the correct result?

To Reproduce

# set OCR_AGENT (I tried both)
os.environ["OCR_AGENT"] = "unstructured.partition.utils.ocr_models.paddle_ocr.OCRAgentPaddle" # use paddle ocr
# os.environ["OCR_AGENT"] = "unstructured.partition.utils.ocr_models.tesseract_ocr.OCRAgentTesseract" # use tesseract
filepath = "Attention Is All You Need.pdf"

raw_pdf = partition_pdf(
    filename=filepath,                  
    strategy="hi_res",             
    infer_table_structure=True,     
    extract_images_in_pdf=True,          
    langueges=['eng'],                  
    chunking_strategy="by_title",        
    max_characters=4000,                
    combine_text_under_n_chars=2000,     
    extract_image_block_types=["Image"], 
    extract_image_block_to_payload=True, 
)
for chunk in raw_pdf:
    for ele in chunk.metadata.orig_elements:
        if "Table" in str(type(ele)):
            print(ele.metadata.text_as_html)

Expected behavior
Let's pick the first result of parsed table, it's supposed to return something like O(n2 . d), O(n . d2), etc in this html table:

<table><thead><tr><th>Layer Type</th><th>Complexity per Layer</th><th>Sequential Operations</th><th>Maximum Path Length</th></tr></thead><tbody><tr><td>Self-Attention</td><td>O(n? - d)</td><td>O(1)</td><td>O(1)</td></tr><tr><td>Recurrent</td><td>O(n - d?)</td><td>O(n)</td><td>O(n)</td></tr><tr><td>Convolutional</td><td>O(k-n-d?)</td><td>O(1)</td><td>O(logx(n))</td></tr><tr><td>Self-Attention (restricted)</td><td>O(r-n-d)</td><td>ol)</td><td>O(n/r)</td></tr></tbody></table>

But instead I got something like O(n? - d), O(n - d?), etc.
('power of 2' parsed as '?' and 'dot multiplication symbol' parsed as '-')

Screenshots
original table in paper:
https://imgur.com/a/2ELXjJ7

Environment Info
Kaggle Env:
Python 3.11.11
unstructured 0.18.1
unstructured-client 0.37.2
unstructured-inference 1.0.5
unstructured_paddleocr 2.10.0
unstructured.pytesseract 0.3.15

Additional context
Add any other context about the problem here.

Start by reproducing the provided partition_pdf call with strategy="hi_res", infer_table_structure=True, and each OCR_AGENT, then inspect metadata.text_as_html for the table elements. Compare the extracted formulas with the linked paper screenshot; done means powers and multiplication symbols in the table are preserved in the generated HTML.